Bayesian Modeling and MCMC Computation in Linear Logistic Regression for Presence-only Data

Ba y esian Mo deling and MCMC Computation in Linear Logistic Regression for Presence-only Data F abio Divino 1 ,Natalia Golini 2 ,Gio v anna Jona Lasinio 2 An tti Pen ttinen 3 1-Division of Ph ysics, Computer Science and Mathematics, Univ ersit y of Molise 2-Departmen t of Statistical Sciences, Univ ersit y of Rome "L a Sapienza" 3-Departmen t of Mathematics and Statistics, Univ ersit y of Jyv äskylä No vem b er 27, 2024 Abstract Presence-only data are referred to situations in which, giv en a censoring mec ha- nism, a b inary resp onse can b e observ ed only with resp ect to on outcome, usually called pr esenc e . In this w ork w e presen t a Ba y esian approach to the problem of presence-only data based on a t w o lev els scheme. A probability law and a case- con trol design are combined to handle the double source of uncertaint y: one due to the censoring and one due to the sampling. W e prop ose a new formalization for the logistic mo del with presence-only data that allo ws further insigh t in to inferen tial issues related to the mo del. W e concen trate on the case of the linear logistic re- gression and, in order to mak e inference on the parameters of interest, w e present a Mark o v Chain Mon te Carlo algorithm with data augmen tation that do es not require the a priori knowledge of the p opulation prev alence. A simulation study concerning 24,000 sim ulated datasets related to different scenarios is presented comparing our prop osal to optimal b enc hmarks. Keyw ords: Ba y esian mo deling, case-control design, data augmentation, logistic regres- sion, Mark ov Chain Monte Carlo, p opulation prev alence, presence-only data, simulation. 1 In tro duction There is a significant b o dy of literature in statistics, econometrics and ecology dealing with the mo deling of discrete resp onses under biased or preferential sampling designs. They are particularly p opular in the natural sciences when sp ecies distributions are studied. Suc h sample schemes may reduce the survey cost esp ecially when one of the resp onses is 1 rare. A large part of statistical literature concerns the case-control design, retrosp ective, c hoice-based or resp onse-based sampling (Lancaster and Imbens, 1996). In the simplest case a sample of cases and a sample of con trols are a v ailable and for each observ ation a set of “attributes/co v ariates” is observ ed in b oth samples. Then inference is carried out follo wing standard statistical pro cedures (Armenian, 2009). A case that has receiv ed increasing atten tion in the literature is the situation where the sample of con trols is a random sample from the whole p opulation with information only on the attributes and not on the resp onse (Lancaster and Imbens, 1996). This situation is fairly common in ecological studies where only sp ecies’ presence is recorded when field surv eys are carried out. In the ecological literature, since the 1990’s such data are called pr esenc e only data (see Araùjo and Williams, 2000, and references therein). P earce and Bo yce (2006) define presence-only data as “consisting only of observ ations of the organism but with no reliable data where the sp ecies was not found”. Atlases, m useum and herbar- ium records, sp ecies lists, incidental observ ation databases and radio-tracking studies are examples of suc h data. In recen t y ears we find a considerably growing literature describing approaches to the mo deling of this type of data, among the man y ecological pap ers w e recall Keating and Cherry (2004), Pearce and Boyce (2006), Elith et al. (2006), Elith and Leath wick (2009), F ranklin (2010) and, most notably , in the statistical literature W ard et al. (2009), W arton and Shepherd (2010), Chakrab ort y et al. (2011), Di Lorenzo et al. (2011) and Dorazio (2012). While in W arton and Shepherd (2010) and Chakrab ort y et al. (2011) to mo del the presence-only data Poisson p oin t pro cesses are considered in the likelihoo d and Ba yesian framew ork respectively , in W ard et al. (2009) and Di Lorenzo et al. (2011) a mo dified case- con trol logistic mo del is adopted in the likelihoo d and Ba y esian p ersp ective resp ectively , in b oth pap ers there is no accoun t for p ossible dep endence structure in the observ ations. In Dorazio (2012) the asymptotic relations b etw een the tw o approac hes are discussed. A differen t approac h, MaxEn t, is based on the maximum entrop y principle (Jaynes, 1957). In MaxEn t (Phillips et al. , 2006; Elith et al. , 2011) the relative entrop y b etw een the dis- tribution of co v ariates at lo cations where the sp ecies is present and the unconditional bac kground distribution of co v ariates is maximized sub ject to some constrains concerning empirical statistics (see Philips et al., 2006, for details). As p oin ted by Dorazio (2012) “the MaxEn t metho d requires kno wledge of sp ecies’ prev alence for its estimator of o ccurrence to b e consisten t”. In what follows we are going to use the name pr esenc e-only data when referring to the ab o v e sketc hed general problem of having information on the presence and co v ariates join tly on a sample from a p opulation, while information on only the cov ariates is av ail- able on an y sample from the same p opulation. This work is dev elop ed in the same discrete setting as in W ard et al. (2009) and Di Lorenzo et al. (2011), i.e., w e hav e a p opulation of indep enden t units, no dep endence structure, such as spatial correlation, is anticipated. W e defer the treatment of this extension to a subsequen t work. The main con tribution of the pap er is a new rigorous formalization of the logistic regres- sion mo del with presence-only data that allows further insight in to the inferen tial issues. This leads us to an algorithmic pro cedure that, among other results, returns a MCMC ap- pro ximation of the resp onse prev alence under general knowledge of the pro cess generating 2 the data. W e also presen t a large sim ulation study in volving 24,000 sim ulated datasets and comparing our approac h to other tw o mo dels representing optimal b enchmark s. The paper is organized as follo ws. Section 2 introduces a general framework for the presence-only data problems, Section 3 presents our Ba y esian approac h, Section 4 de- scrib es the MCMC algorithm while results related to the simulation study are rep orted in Section 5. Finally in Section 6 some conclusions are drawn and future dev elopmen ts briefly describ ed. 2 Linear logistic regression for presence-only data. The analysis of a binary resp onse related to a set of explicative cov ariates is usually carried out through the use of the logistic regression where the logit of the conditional probabilit y of occurrence is mo deled as a function of cov ariates. In this section, we first in tro duce a general framework for the mo deling of presence-only data and then consider the case of the linear logistic regression. The approac h prop osed is built on tw o levels and w e partially follo w the form ulation in tro duced b y W ard et al. (2009) but adopting a Ba y esian scheme as in Divino et al. (2011). 2.1 A t wo level approach. Let Y b e a binary v ariable informing on the presence ( Y = 1 ) or absence ( Y = 0 ) of a p opulation’s attribute and let X = ( X 1 , ..., X k ) denote a set of highly informative, on the same attribute, cov ariates which are av ailable on the same p opulation. Then, the presence-only problem can b e formalized b y considering a censorship mec hanism that acts when observing the resp onse Y , so that part of the p opulation units are not reac hable. In particular, we refer to the situation in which we are able to detect only a partial set of units on whic h the attribute of interest is present while the information on the co v ariates X is a v ailable on the en tire p opulation. In this situation, we hav e to consider tw o t yp es of uncertain t y: the uncertaint y due to the mechanism of censorship and the uncertain t y due to the sampling pro cedure. Moreo ver, since we are not able to collect a random sample of observ able data, w e need to adjust for the sampling mec hanism through the use of a case-con trol sc heme (Breslow and Dey, 1980; Breslow, 2005; Armenian, 2009) . In order to build a Bay esian mo del, in this framew ork we adopt the following conceptual sc heme in tw o levels. Lev el 1. Giv en the p opulation of interest U of size N , the binary resp onses y = ( y 1 , ..., y N ) are generated indep enden tly by a probability law M . Lev el 2. Let U p b e the subset of U where w e observe Y = 1 . A mo dified case-control design is applied so that a sample of presences, considered as cases, is selected from U p 3 and a sample of “contaminated” controls (Lancaster and Im b ens, 1996) is selected from the whole p opulation U , with all the cov ariates but no information on Y . Here, w e cannot approac h the mo del construction using only a finite p opulation ap- proac h (Särndal, 1978) b ecause of the censoring mec hanism that “masks” distributional information on Y already at the p opulation level. By the introduction of Level 1 we can describ e the censored observ ations as random quantities generated by the mo del M . Hence, the problem of presence-only data can b e formalized as a problem of missing data (Rubin, 1976; Little and Rubin, 1987). 2.2 The mo del generating p opulation data. A t the first lev el, w e assume that the law M is defined in terms of the conditional probabilit y of o ccurrence P r ( Y = 1 | x ) , denoted b y π ∗ ( x ) , when the co v ariates are X = x . Moreo v er, w e consider that the relation b et ween Y and X is formalized through a regression function φ ( x ) on the logit scale φ ( x ) = logit π ∗ ( x ) , (1) that is π ∗ ( x ) = e φ ( x ) 1 + e φ ( x ) . (2) When the data y = ( y 1 , ..., y N ) are indep endently generated from M , we denote b y π the empirical prev alence of the binary resp onse Y in U , expressed as the ratio of the n um b er of presences N 1 to the size of the p opulation, that is π = N 1 N . 2.3 The mo dified case-con trol design. A t the second level, w e adopt a case-control design mo dified for presence-only data (Lan- caster and Imbens, 1996) in order to account for the sp ecific sampling pro cedure consid- ered. The use of the case-con trol scheme is necessary at all times when it is appropriate to select observ ations in fixed prop ortions with resp ect to the v alues of the resp onse v ari- able. This can o ccur when the attribute of interest represen ts a phenomenon that is rare among the units of the p opulation as for example a rare disease or a rare exp osure in epidemiological studies (W o o dward, 2005). No w, let C b e a binary indicator of inclusion into the sample ( C = 1 denotes that a unit is in the sample), let ρ 0 = P r ( C = 1 | Y = 0) and ρ 1 = P r ( C = 1 | Y = 1) b e the inclusion probabilit y of the absences and the presences, resp ectiv ely . Under the assumption that, giv en Y , the sampling mechanism is indep endent from the cov ariates X , the conditional probabilit y of o ccurrence is mo dified through the Bay es rule as P r ( Y = 1 | C = 1 , x ) = ρ 1 e φ ( x ) ρ 0 + ρ 1 e φ ( x ) . (3) 4 Hence, the corresp onding case-con trol regression function φ cc ( x ) defined as the logit of (3) is giv en by φ cc ( x ) = φ ( x ) + log ρ 1 ρ 0 . (4) In particular, if the selection of cases ( n 1 ) and controls ( n 0 ) is made indep endently without replacemen t, the inclusion probabilities are given in terms of the empirical prev alence π b y ρ 0 = n 0 (1 − π ) N and ρ 1 = n 1 π N , so that the equation (4) b ecomes φ cc ( x ) = φ ( x ) + log n 1 n 0 − log π 1 − π . (5) In our framew ork, since the resp onse v ariable Y is already censored at the p opulation lev el, the standard case-control design cannot b e adopted but it should b e mo dified in suc h a w a y that a sample of presences is matched with an indep endent sample drawn from the entire p opulation, named the b ackgr ound sample (Zaniewski et al. , 2002; W ard et al. , 2009). Remark that in this sample the resp onse v ariable is unobserv ed and only the co v ariates are av ailable. In this w ay , the complete sample S is comp osed by a set S u of n u indep enden t background data, where the resp onse Y is not observ ed, drawn from the en tire U and b y a set S p of n p indep enden t observ ations selected from the sub-p opulation of presences U p . This pro cedure implies that the reference p opulation U is augmented with its subset U p so that the total n umber of observ ations considered in the sampling scheme b ecomes N + N 1 . T o illustrate the sampling framework we are going to adopt here, let us consider the follo wing situation: w e can lab el p opulation units of type y = 1 only when they are isolated from units of t yp e y = 0 . This can b e formalized b y introducing a binary stratum v ariable Z suc h that Z = 0 indicates when an observ ation is dra wn from the en tire p opulation U while Z = 1 denotes the sampling from the sub-p opulation U p . Remark that Z = 1 implies Y = 1 whilst Z = 0 implies that Y is an unknown v alue y ∈ { 0 , 1 } . Moreo v er, b y construction Z is indep endent from the co v ariates X , given the resp onse Y . The in tro duction of the v ariable Z allo ws us to define the structure of the data at the p opulation lev el and at the sample level in terms of presences/absences ( Y ) and kno wn/unkno wn data ( Z ), as rep orted in T able 1 and T able 2. Y/Z Z = 0 Z = 1 T otal Y = 0 N 0 0 N 0 Y = 1 N 1 N 1 2 N 1 T otal N N 1 N + N 1 T able 1: Data structure at the p opulation level. 5 Y/Z Z = 0 Z = 1 T otal Y = 0 n 0 u 0 n 0 Y = 1 n 1 u n p n 1 T otal n u n p n T able 2: Data structure at the sample lev el. In T able 1, N 0 is the n um b er of absences in the p opulation U while in T able 2, n 0 u and n 1 u resp ectiv ely denote the unkno wn frequencies of absences and presences in the sub-sample S u . Remark that, in the ab ov e describ ed situation, the inclusion probability of units with or without the men tioned attribute changes. In fact, while an absence can b e drawn only when sampling from U , a presence can b e selected when sampling b oth from U and from U p . Th us, one has ρ 0 = n 0 N 0 = n 0 u (1 − π ) N , (6) and ρ 1 = n 1 2 N 1 = n 1 u + n p 2 π N . (7) The introduction of the stratum v ariable Z allo ws us also to exactly derive the logis- tic regression mo del under the case-con trol design mo dified for presence-only data. In fact, when w e consider the p opulation U augmen ted with its subset U p , the mo del π ∗ ( x ) represen ts the conditional probabilit y to mark a presence only when Z = 0 , that is P r ( Y = 1 | Z = 0 , x ) = π ∗ ( x ) . On the other hand, when Z = 1 , w e simply hav e P r ( Y = 1 | Z = 1 , x ) = 1 . W e can prov e the follo wing result. Prop osition 1. Under the assumption that Z is indep enden t from X given Y , one has P r ( Y = 1 | x ) = 2 π ∗ ( x ) 1 + π ∗ ( x ) . (8) Pro of. F rom the h yp othesis of conditional indep en dence it results P r ( Z | Y , x ) = P r ( Z | Y ) , that can b e express also as P r ( Y | Z , x ) P r ( Z | x ) P r ( Y | x ) = P r ( Y | Z ) P r ( Z ) P r ( Y ) . Let consider the case with Y = 1 and Z = 0 , one has P r ( Y = 1 | Z = 0 , x ) P r ( Z = 0 | x ) P r ( Y = 1 | x ) = P r ( Y = 1 | Z = 0) P r ( Z = 0) P r ( Y = 1) . 6 The probabilities enclosed in the second term can b e derived from T able 1 and one has π ∗ ( x ) P r ( Z = 0 | x ) P r ( Y = 1 | x ) = N 1 N N N + N 1 2 N 1 N + N 1 = 1 2 . (9) In the case Y = 1 and Z = 1 one similarly obtains P r ( Z = 1 | x ) P r ( Y = 1 | x ) = N 1 N 1 N 1 N + N 1 2 N 1 N + N 1 = 1 2 . (10) F rom (10) it results P r ( Y = 1 | x ) = 2 P r ( Z = 1 | x ) and b y substituting in (9), one can deriv e that P r ( Z = 0 | x ) = 1 1 + π ∗ ( x ) and hence P r ( Z = 1 | x ) = π ∗ ( x ) 1 + π ∗ ( x ) . Now, it is simple to obtain that P r ( Y = 1 | x ) = 2 π ∗ ( x ) 1 + π ∗ ( x ) .  If w e assume that, giv en Y , the inclusion in to the sample ( C = 1 ) is indep endent from the co v ariates X , one has 1 P r ( Y = 0 | C = 1 , x ) P r ( C = 1 | x ) = 1 − π ∗ ( x ) 1 + π ∗ ( x ) ρ 0 (11) and P r ( Y = 1 | C = 1 , x ) P r ( C = 1 | x ) = 2 π ∗ ( x ) 1 + π ∗ ( x ) ρ 1 . (12) Then, from the ratio of (12) to (11), it results P r ( Y = 1 | C = 1 , x ) P r ( Y = 0 | C = 1 , x ) = 2 π ∗ ( x ) 1 − π ∗ ( x ) ρ 1 ρ 0 , and b y plugging the quan tities ρ 0 and ρ 1 , as defined in (6) and in (7), in to the logit of P r ( Y = 1 | C = 1 , x ) , one obtains the following relation logit P r ( Y = 1 | C = 1 , x ) = log  2 π ∗ ( x ) 1 − π ∗ ( x ) ρ 1 ρ 0  = log  2 π ∗ ( x ) 1 − π ∗ ( x ) n 1 u + n p n 0 u 1 − π 2 π  = log  π ∗ ( x ) 1 − π ( x ) n 1 u + n p n 0 u 1 − π π  = logit π ∗ ( x ) + log n 1 u + n p n 0 u − log π 1 − π = φ ( x ) + log n 1 u + n p n 0 u − log π 1 − π , (13) 1 see Appendix for the detailed pro of. 7 that represen ts the logistic regression mo del under the case-control design for presence- only data. As w ell, we can no w formalize the presence-only data regression function φ pod ( x ) as φ pod ( x ) = φ ( x ) + log n 1 u + n p n 0 u − log π 1 − π . (14) Although the deriv ation is substantially different, w e end with the same form ulation as in W ard et al. (2009). No w, in order to mak e parameter estimation p ossible, we need to handle the ratio ρ 1 ρ 0 = n 1 u + n p n 0 u 1 − π 2 π , (15) where the quan tities π and n 1 u are unkno wn ( n 0 u = n u − n 1 u ). In the recent literature, tw o main approaches ha v e b een prop osed. The first one b y W ard et al. (2009) replace the ratio n 1 u + n p n 0 u with the ratio of the exp ected num b ers of presences and absences in the sample, that is ρ 1 ρ 0 ≈ E [ n 1 u + n p ] E [ n 0 u ] 1 − π 2 π = π n u + n p (1 − π ) n u 1 − π 2 π = π n u + n p 2 π n u . (16) These authors adopt a lik eliho o d approach and computation is carried out via the EM al- gorithm. As they underline, this appro ximation can b e easily implemen ted if the empirical p opulation prev alence π is known a priori. They discuss also the p ossibility to estimate π join tly with the regression function when the prev alence is identifiable, as for example in the linear logistic regression, and with resp ect to this case they presen t a simulation example. The difficult y in obtaining efficien t joint estimates b ecause of the correlation b et w een π and the intercept of the linear regression term is discussed. Notice that W ard et al. (2009) considers a slightly differen t representation of the ratio (16), omitting the m ultiplier “2” in the denominator. Di Lorenzo et al. (2011), dealing with a problem of abundance data, use the approximation (16), but they adopt a B a yesian approach and consider the p opulation prev alence π as a further parameter in the mo del. They choose an informative Beta prior for π , but their MCMC algorithm contains an un usual w eakness since the simulation of π is p erformed from its prior and not from the p osterior that can b e derived through the interaction b et w een the parameter π and the regression function φ ( x ) . A differen t appro ximation of the ratio (15) can b e obtained by considering the sample prev alence in S u (the bac kground sample) π u = n 1 u n u , where n 1 u = X i ∈ S u y i . Due to the censorship pro cess, this quantit y is unkno wn but it w ould b e the maximum lik eliho o d estimator for π if the data y u = { y i , i ∈ S u } could b e observed. Now, replacing π by π u in (15) one obtains ρ 1 ρ 0 ≈ n 1 u + n p n 0 u 1 − π u 2 π u = n 1 u + n p n u − n 1 u n u − n 1 u 2 n 1 u = n 1 u + n p 2 n 1 u , (17) 8 that allo ws to formulate a computable version of the regression function for presence-only data as φ pod ( x ) ≈ φ ( x ) + log n 1 u + n p n 1 u . (18) This function dep ends on the data y u in S u whic h are not directly observ able, but if y u is treated as missing data one can enclose it in to the estimation pro cess and then obtain a consistent approximation for φ pod ( x ) . In particular, in a Bay esian framew ork, this idea can be p erformed by using a Mark o v Chain Mon te Carlo computation with data augmen tation. Moreo v er from the use of MCMC sim ulations we can also obtain an appro ximation of π u and therefore an estimate of the empirical p opulation prev alence π . Details are giv en in Section 4. The approximation (17) can, in principle, b e alwa ys adopted, but some care must be used as iden tifiability issues are present. W e follo w the recommendation in W ard et al. (2009) to approach join tly estimates of φ ( x ) and π only when the latter is identifiable with resp ect to the regression function, as for example in the linear regression case (see W ard et al. , 2009, for mathematical details). 2.4 The linear logistic regression. If w e consider a linear regression function φ ( x ) = xβ , where β = ( β 1 , ..., β k ) is the vector of the regression parameters, a computable model for presence-only data can b e defined through the follo wing approximation φ pod ( x ) ≈ xβ + log n 1 u + n p n 1 u , (19) or equiv alently through the approximation of the conditional probability of o ccurrence at the sample lev el P r ( Y = 1 | C = 1 , x ) ≈ exp { xβ + log n 1 u + n p n 1 u } 1 + exp { xβ + log n 1 u + n p n 1 u } =  1 + n p n 1 u  exp { xβ } 1 +  1 + n p n 1 u  exp { xβ } . (20) In this particular case, all the unkno wns of the mo del are the linear parameters v ector β and the missing data y u in the bac kground sample S u . 3 The hierarc hical Ba y esian mo del. Due to the censorship pro cess affecting the data, we can acquire complete information only on the stratum v ariable Z and not on the binary resp onse Y . Then, it seems natural to mo del Z as the observ able v ariable. If w e consider the conditional join t distribution of Z and Y P r ( Z, Y | C = 1 , x ) = P r ( Z | Y , C = 1 , x ) P r ( Y | C = 1 , x ) , (21) 9 through the marginalization ov er Y , the probabilit y P r ( Z | C = 1 , x ) can b e obtained and w e can express the relation b et ween presences and cov ariates in terms of regression of Z resp ect to X . Notice that, while P r ( Y | C = 1 , x ) can b e obtained from (20), the term P r ( Z | Y , C = 1 , x ) , due to the conditional independence b etw een Z and X giv en Y , simply reduce to b e equal to P r ( Z | Y , C = 1) that can b e derived from T able 2. W e p oint out that, even if the resp onse Y do es not play an explicit role after the marginal- ization step, we need to keep it in the mo del as a hidden v ariable in order to obtain the appro ximation for the quan tit y n 1 u = P i ∈ S u y i , necessary to correct the linear regression function for presence-only data. No w, w e can formalize the hierarchical Ba y esian mo del to estimate the parameters of a linear logistic regression under the case-control scheme adjusted for presence-only data. In order to b etter explain the conditional relationship underlying the hierarch y , w e in tro- duce the graph in Figure 1. The dashed no de indicates a v ariable hidden with resp ect to the conditional relationships. θ β Y u Y p Z C = 1 X Figure 1: Graphical represen tation of the hierarchical Bay esian mo del. The priors. A t the top of the hierarch y , w e assume the hyper parameter θ distributed as p ( θ ) . A t the second lev el, w e consider the prior probability distribution on β dep ending 10 on the h yp er parameter θ , that is β | θ ∼ p ( β | θ ) . At the third level, the unobserved data y u in S u are considered laten t parameters with prior distribution Bernoul li (denoted b y B e ) with probabilit y of o ccurrence given by the approximation in (20), that is y i | C i = 1 , x i , β ∼ B e    1 + n p n 1 u  exp { xβ } 1 +  1 + n p n 1 u  exp { xβ }   , i ∈ S u . This p oint is imp ortant for deriving the predictive distribution of the unobserved data y u necessary in the estimation algorithm. The lik eliho o d. At the low est lev el of the hierarch y , w e hav e the likelihoo d, defined with resp ect to the observ able stratum v ariable Z . Recalling that from the T able 2 we ha v e P r ( Z = 1 | Y = 0 , C = 1) = 0 and P r ( Z = 1 | Y = 1 , C = 1) = n p n 1 u + n p , when (21) is marginalized o ver Y , one obtains the appro ximation P r ( Z = 1 | C = 1 , x, β ) ≈ n p n 1 u + n p exp { xβ + log n 1 u + n p n 1 u } 1 + exp { xβ + log n 1 u + n p n 1 u } = n p n 1 u exp { xβ } 1 +  1 + n p n 1 u  exp { xβ } (22) and hence P r ( Z = 0 | C = 1 , x, β ) = 1 − P r ( Z = 1 | C = 1 , x, β ) ≈ 1 + exp { xβ } 1 +  1 + n p n 1 u  exp { xβ } . (23) Th us, w e can assume that for all i ∈ S the conditional distribution of Z i is Bernoul li with probabilit y of o ccurence giv en by (22), that is Z i | C i = 1 , x i , β ∼ B e   n p n 1 u exp { xβ } 1 +  1 + n p n 1 u  exp { xβ }   , i ∈ S. Recalling that Z i = 0 for all i ∈ S u while Z i = 1 for all i ∈ S p , the likelihoo d function can b e written as L ( β ; z , x ) = Y i ∈ S u 1 + exp { x i β } 1 +  1 + n p n 1 u  exp { x i β } × Y i ∈ S p n p n 1 u exp { x i β } 1 +  1 + n p n 1 u  exp { x i β } . W ard et al. (2009) defines this function as the observe d likeliho o d versus the ful l likeliho o d that, instead, considers the distribution of the stratum v ariable Z jointly with the resp onse Y . 11 The p osterior. No w, through the Ba yes rule we derive the full p osterior p ( β , θ | z , x ) ∝ p ( θ ) p ( β | θ ) L ( β ; z , x ) (24) that can b e used to make inference on the quantities of interest. 4 The MCMC computation. Samples from (24) can b e obtained via Marko v Chain Mon te Carlo simulation (Rob ert and Casella, 2004; Liu, 2008). While it seems quite standard to implement a direct sam- pler for the v ector β and the hyper parameter θ , we need to sample also the laten t y u . F or this reason we introduce a step of data augmentation (T anner and W ong, 1987; T anner, 1996) in the estimation pro cedure. The basic idea of the data augmentation tec hnique is to augment the set of observ ed data to a set of completed data that follo w a simpler distribution (Liu and W u, 1999). In our framework, w e need to augmen t the observ ations of the stratum v ariable z with the missing v alues y u in order to hav e, at eac h itera- tion, a consistent v alue of the quantit y n 1 u , necessary to adjust the regression function φ pod ( x ) ≈ xβ + log n 1 u + n p n 1 u for presence-only data. The following result allo ws for an easy implemen tation of the data augmentation step. Prop osition 2. Using the approximation (17) of the ratio (15), the p osterior predictive probabilit y of o ccurrence for an unobserv ed response y in the sub-sample S u is approxi- mated b y the mo del M that generates the data at the p opulation level, that is P r ( Y = 1 | Z = 0 , C = 1 , x ) ≈ π ∗ ( x ) . (25) Pro of. F rom the conditional indep endence b etw een Z and X given Y , the predictiv e probabilit y of o ccurrence in S u is giv en by P r ( Y = 1 | Z = 0 , C = 1 , x ) = P r ( Z = 0 | Y = 1 , C = 1) P r ( Y = 1 | C = 1 , x ) P r ( Z = 0 | C = 1 , x ) . F rom T able 2 w e ha ve that P r ( Z = 0 | Y = 1 , C = 1) = n 1 u n p + n 1 u and hence P r ( Y = 1 | Z = 0 , C = 1 , x ) = n 1 u n p + n 1 u P r ( Y = 1 | C = 1 , x ) P r ( Z = 0 | C = 1 , x ) . (26) No w, recalling that in the general case one has P r ( Y = 1 | C = 1 , x ) ≈  1 + n p n 1 u  exp { φ ( x ) } 1 +  1 + n p n 1 u  exp { φ ( x ) } (27) 12 and P r ( Z = 0 | C = 1 , x ) ≈ 1 + exp { φ ( x ) } 1 +  1 + n p n 1 u  exp { φ ( x ) } , (28) b y substituting (27) and (28) in (26), one obtains P r ( Y = 1 | Z = 0 , C = 1 , x ) ≈ n 1 u n p + n 1 u  1 + n p n 1 u  exp { φ ( x ) } 1 + exp { φ ( x ) } = exp { φ ( x ) } 1 + exp { φ ( x ) } = π ∗ ( x ) .  4.1 The data augmen tation algorithm. A general MCMC scheme to p erform inference on a linear regression mo del for presence- only data can b e defined as follow. Step 0. Initialize θ , β and y u Step 1. Set n 1 u = P i ∈ S u y i Step 2. Sample θ from p ( θ | z , x , β ) Step 3. Sample β from p ( β | z , x , θ ) Step 4. Sample y i from p ( y i | Z i = 0 , C i = 1 , x i , β ) for all i ∈ S u Goto Step 1 After the initialization of all the arra ys (Step 0), Step 1 sets a curren t v alue for the quantit y n 1 u to adjus t the regression function φ pod ( x ) . Step 2 and Step 3 consider the sampling from the p osterior of the h yp er parameter θ and the regression parameter β , resp ectiv ely , and they can b e p erformed by Metrop olis-Hasting schemes (Rob ert and Casella, 2004). Step 4 concerns the data augmentation for the unobserv ed y u in order to up date consistently the quan tit y n 1 u at the following iteration. F rom the result (25), this simulation can b e obtained b y a Gibbs sampler (Rob ert and Casella, 2004) since the posterior predictiv e distribution for all i ∈ S u is approximated b y Bernoul li with parameter of occurrence π ( x i ) = exp { x i β } 1+exp { x i β } . 13 4.2 The estimation of the prev alence π . F rom the data augmentation algorithm w e can obtain a MCMC estimate of the p opulation prev alence π . In fact, if at eac h iteration t , after the Marko v chain has reached the equi- librium, w e sav e the current v alue n ( t ) 1 u , w e can obtain a consistent MCMC approx imation of the sample prev alence π u in S u b y ˆ π mcmc = ¯ n 1 u n u (29) where ¯ n 1 u is the ergo dic mean of the augmentations n ( t ) 1 u o v er the Marko v chain, that is ¯ n 1 u = P T t =1 n ( t ) 1 u T . Therefore, since π u w ould b e a consisten t estimator for π , ˆ π mcmc represen ts also a consis- ten t estimation of the empirical p opulation prev alence. 5 A comparativ e sim ulation study . W e presen t a sim ulation exp eriment to ev aluate the p erformances of the mo del (20). T o this aim we generate several datasets in the w a y describ ed b elow and we compare our prop osal with resp ect to tw o mo dels acting in tw o different situations: (a) the censorship pro cess do es not act on the p opulation U so that the data y are completely observ ed; (b) the censorship is presen t, but we assume kno wn the p opulation prev alence so that appro ximation (16) can b e used. In (a) w e are able to estimate a linear logistic mo del (denoted by M 0 ), no correction is required and φ 0 ( x ) = xβ . In (b) w e consider a linear logistic mo del for presence-only data, denoted by M 1 , with regression function φ 1 ( x ) = xβ + π n u + n p π n u . Mo del (20) (denoted by M 2 ) is estimated when the censorship pro cess acts on the data and no information is av ailable on the p opulation prev alence. In this case, the regression function is giv en by φ 2 ( x ) = xβ + n 1 u + n p n 1 u . Remark that mo del M 2 can b e estimated when the least amount of information is a v ailable, M 1 requires less information than M 0 but more than M 2 and M 0 can b e used only in the ideal s ituation of complete information. W e assume M 1 as b enc hmark mo del in the case of presence-only data. The generation of data. In order to set the simulation study , we need to generate the co v ariates X and the binary resp onse Y . In particular, we consider tw o cov ariates: X 1 , giving strong information on the distribution of the resp onse Y , and X 2 , representing a term of noise, not av ailable in the estimation step. W e assume X 1 distributed as a mixture of t w o Gaussian densities (denoted by N ), centred in µ a = 4 . 0 and µ b = − 4 . 0 resp ectiv ely , and with equal v ariances σ 2 = 4 . 0 , that is X 1 ∼ w N a ( µ a ; σ 2 ) + (1 − w ) N b ( µ b ; σ 2 ) . The weigh t w is a realization of a Bernoul li random v ariable with probability of o ccurrence fixed to p = 0 . 165 . X 2 has standard Gaussian distribution N (0 , 1) . Finally , the binary 14 resp onse Y , given the cov ariates X 1 and X 2 , is Bernoul li distributed with probability of o ccurrence π ( x ) = exp { β 0 + β 1 x 1 + β 2 x 2 } 1 + exp { β 0 + β 1 x 1 + β 2 x 2 } . W e generate co v ariates and binary response with resp ect to a population U of size N = 10000 . Three general scenarios with differen t level of complexity hav e b een considered: (i) β 0 = 0 , β 1 = 1 , β 2 = 0 : only the informative co v ariate X 1 generates the data; (ii) β 0 = 0 , β 1 = 1 , β 2 = 1 : a term of noise X 2 is added to the informative co v ariate; (iii) β 0 = 1 , β 1 = 1 , β 2 = 1 : X 1 , X 2 and a constant effect generate the data. The case-con trol sampling. F or each scenario, we sample under the case-control de- sign with a ratio of presence/unobserv ed equal to 1 : 4 and with resp ect to eight different sample sizes n = 50 , 100 , 200 , 500 , 1000 , 1500 , 2000 , 3000 . F or example, if the sample size is equal to n = 500 , we build the corresp onding simulated exp erimen t by extracting a rand om sample S p from U p of n p = 100 presences and a random sample S u from U of n u = 400 unobserved v alues, co v ariates are av ailable for the whole sample S . W e consider k = 1000 indep endent replications of eac h exp erimen t. In summary , w e generate a database of 24,000 datasets (8 sample sizes, 3 scenarios and 1000 replications). With resp ect to the generating of the data, we considered a quite general framework since the contribution of an informativ e cov ariate was combined with a constan t effect and a white Gaussian noise. With resp ect to the three scenarios, w e obtained empirical p opulation prev alences resp ectiv ely π ( i ) = 0 . 215 , π ( ii ) = 0 . 223 and π ( iii ) = 0 . 286 . The MCMC estimation. The estimation is p erformed in a Ba yesian framew ork for all the mo dels M 0 , M 1 and M 2 . The likelihoo d function we use in the estimation is based on a mo del that do es not alw a ys replicate the mo del used to generate data. More precisely for all exp erimen ts (i), (ii) and (iii) the estimation mo del is: logit ( P r ( Y = 1 | X 1 = x 1 )) = β 0 + β 1 x 1 (30) than with scenario (i) the mo del that generates the data and the one defining the likelihoo d are the same, whilst for scenarios (ii) and (iii) the likelihoo d mo del b ecomes increasingly differen t from the one that generates the data. Notice that w e consider a simpler structure than the one shown in Figure (1) as we choose a Gaussian prior N (0 , 25) for all regression parameters ( β 0 , β 1 ) and no h yp er parameter is considered. Then, MCMC estimates are computed using 5000 runs after 10000 iterations of burn-in, no thinning is applied as samples auto correlation is negligible. 15 Results. In what follows we rep ort Figures and T ables built on scenario (iii) as it rep- resen ts the most complex of the three alternativ es and it is our “w orst” case. In each replicate of an exp erimen t, p oin t estimates are computed as p osterior means ov er 5000 iterations. In Figure 2 b oxplots describing p oint estimates b eha viour are rep orted, hor- izon tal lines corresp onding to the “true” v alues are dra wn. The first b ox corresp onds to pro cedure M 0 , the second to M 1 and the third to our prop osal M 2 . In M 0 the prev alence π is estimated as the ratio of the observed presences in S u to the sample size n u . In M 1 , although π is assumed kno wn a priori, w e consider its p osterior prediction in S u . Finally in M 2 , the prev alence is obtained at each MCMC step as describ ed in section 4.2 and then the mean o v er 5000 runs is taken. In T able 3 further details of the p oin t estimates are rep orted: the median and in paren thesis the first and third quartiles. F rom the Figures and the v alues w e can see that the three pro cedure lead to “comparable” v alues with the ob vious reduction of v ariability when n increases. Remark that the estimates for M 2 , although affected by a larger v ariability with small sample sizes, rapidly approaches M 0 and M 1 b eha viour with increasing sample size. This can b e seen more clearly in Figure 3 where ro oted mean square errors (rmse) are rep orted. As far as β 1 is concerned the lac k of knowledge on X 2 leads to biased p oin t estimates regardless the estimation pro cedure. T ables 5 and 6 in App endix rep ort p oint estimates for scenarios (i) and (ii). F or scenarios (i) unbiased estimates are obtained while (ii) is affected by the s ame distortion as (iii) but with smaller v ariability . 16 n Model β 0 β 1 π 50 M 0 1.42 (0.68 ; 2.33) 1.15 (0.88 ; 1.55) 0.28 (0.25 ; 0.35) M 1 3.13 (1.78 ; 4.46) 1.69 (1.17 ; 2.28) 0.31 (0.26 ; 0.35) M 2 1.79 (-3.38 ; 4.26) 1.44 (0.76 ; 2.12) 0.24 (0.13 ; 0.34) 100 M 0 1.14 (0.72 ; 1.62) 1.00 (0.86 ; 1.22) 0.29 (0.25 ; 0.33) M 1 2.12 (1.11 ; 3.51) 1.30 (0.97 ; 1.80) 0.30 (0.26 ; 0.34) M 2 1.92 (0.16 ; 3.87) 1.24 (0.89 ; 1.78) 0.28 (0.21 ; 0.36) 200 M 0 1.01 (0.72 ; 1.36) 0.94 (0.83 : 1.06) 0.29 (0.26 ; 0.31) M 1 1.53 (0.89 ; 2.39) 1.08 (0.87 ; 1.37) 0.29 (0.27 ; 0.32) M 2 1.49 (0.59 ; 2.62) 1.07 (0.83 ; 1.37) 0.29 (0.24 ; 0.34) 500 M 0 0.94 (0.75 ; 1.15) 0.89 (0.82 ; 0.96) 0.29 (0.27 ; 0.30) M 1 1.12 (0.78 ; 1.57) 0.94 (0.82 ; 1.10) 0.29 (0.28 ; 0.31) M 2 1.17 (0.62 ; 1.82) 0.94 (0.80 ; 1.12) 0.29 (0.26 ; 0.32) 1000 M 0 0.91 (0.78 ; 1.04) 0.88 (0.83 ; 0.92) 0.28 (0.28 ; 0.30) M 1 1.03 (0.79 ; 1.34) 0.91 (0.83 ; 1.01) 0.29 (0.28 ; 0.30) M 2 1.05 (0.68 ; 1.49) 0.91 (0.82 ; 1.03) 0.29 (0.27 ; 0.31) 1500 M 0 0.89 (0.80 ; 1.00) 0.86 (0.83 ; 0.91) 0.29 (0.28 ; 0.29) M 1 1.00 (0.78 ; 1.24) 0.89 (0.82 ; 0.98) 0.29 (0.28 ; 0.30) M 2 1.01 (0.71 ; 1.35) 0.90 (0.82 ; 0.99) 0.29 (0.27 ; 0.31) 2000 M 0 0.89 (0.82 ; 0.98) 0.87 (0.84 ; 0.90) 0.29 (0.28 ; 0.29) M 1 0.96 (0.79 ; 1.15) 0.89 (0.83 ; 0.95) 0.29 (0.28 ; 0.29) M 2 0.96 (0.71 ; 1.23) 0.88 (0.82 ; 0.96) 0.29 (0.27 ; 0.30) 3000 M 0 0.90 (0.83 ; 0.97) 0.87 (0.84 ; 0.89) 0.29 (0.28 ; 0.29) M 1 0.94 (0.82 ; 1.09) 0.88 (0.84 ; 0.93) 0.29 (0.28 ; 0.29) M 2 0.95 (0.76 ; 1.17) 0.88 (0.83 ; 0.94) 0.29 (0.28 ; 0.30) T able 3: Scenario (iii): p oin t estimates of regression parameters and prev alence computed as medians o v er 1000 replicates with increasing sample sizes and different mo dels ( M 0 , M 1 and M 2 ). In paren thesis distributions quartiles are rep orted. 17 n β 0 β 1 π 50 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 −10 −5 0 5 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 0 2 4 6 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 0.0 0.2 0.4 0.6 0.8 100 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 −10 −5 0 5 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 0 2 4 6 ● ● ● ● ● ● ● ● ● ● M0 M1 M2 0.0 0.2 0.4 0.6 0.8 200 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 −10 −5 0 5 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 0 2 4 6 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 0.0 0.2 0.4 0.6 0.8 500 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 −10 −5 0 5 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 0 2 4 6 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 0.0 0.2 0.4 0.6 0.8 1000 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 −10 −5 0 5 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 0 2 4 6 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 0.0 0.2 0.4 0.6 0.8 1500 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 −10 −5 0 5 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 0 2 4 6 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 0.0 0.2 0.4 0.6 0.8 2000 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 −10 −5 0 5 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 0 2 4 6 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 0.0 0.2 0.4 0.6 0.8 3000 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 −10 −5 0 5 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 0 2 4 6 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● M0 M1 M2 0.0 0.2 0.4 0.6 0.8 Figure 2: Scenario (iii): b oxplots of simulations with in creasing sample sizes and differen t mo dels( M 0 , M 1 and M 2 ). 18 ● ● ● ● ● ● ● ● 0 500 1000 1500 2000 2500 3000 0 1 2 3 4 sample size rmse ● M0 M1 M2 (a) ● ● ● ● ● ● ● ● 0 500 1000 1500 2000 2500 3000 0.2 0.4 0.6 0.8 1.0 1.2 sample size rmse ● M0 M1 M2 (b) ● ● ● ● ● ● ● ● 0 500 1000 1500 2000 2500 3000 0.05 0.10 0.15 sample size rmse ● M0 M1 M2 (c) Figure 3: Scenario (iii): ro ot mean squared errors for differen t mo dels ( M 0 , M 1 and M 2 ) o v er the 1000 replications, plots with increasing sample sizes for β 0 (a), β 1 (b) and π (c). Dashed tra jectories are rep orted to sho w the patterns. 19 F rom W ard et al. (2009) we kno w that pairwise correlation b et w een parameters is presen t. In T able 4 we rep ort the empirical pairwise correlation measures, obtained as the av erages with resp ect to the 1000 samples, with increasing sample sizes across the differen t mo dels. No significant differences in the pattern of correlation ( β 0 ; β 1 ) b etw een the mo dels M 1 and M 2 while the correlation ( β 1 , π ) has a general weak er pattern in M 1 than M 2 . With respect to the correlation ( β 0 , π ) more significan t difference are present b et w een M 1 and M 2 . In Figure 4 scatterplots of β 0 v ersus π in the 1000 replicates are plotted with equal axis across estimation pro cedures and sample sizes. These pictures help us to understand how this correlation evolv es with increasing sample sizes. M 2 pro duces the most p ositiv e correlated p oin t estimates this b eing an adv antage whenever the mo del is prop erly sp ecified. Mo del M 0 M 1 M 2 n β 0 ; β 1 β 0 ; π β 1 ; π β 0 ; β 1 β 0 ; π β 1 ; π β 0 ; β 1 β 0 ; π β 1 ; π 50 0.65 0.26 -0.09 0.59 0.10 -0.30 0.68 0.81 0.31 100 0.75 0.24 -0.12 0.89 0.29 0.02 0.82 0.76 0.37 200 0.78 0.34 -0.04 0.94 0.39 0.18 0.90 0.78 0.48 500 0.79 0.38 0.00 0.95 0.41 0.24 0.92 0.77 0.51 1000 0.77 0.38 -0.01 0.94 0.46 0.27 0.91 0.81 0.54 1500 0.78 0.42 0.00 0.95 0.48 0.28 0.92 0.81 0.55 2000 0.77 0.35 -0.06 0.94 0.49 0.30 0.92 0.81 0.55 3000 0.81 0.37 -0.01 0.95 0.43 0.23 0.91 0.80 0.52 T able 4: Scenario (iii): pairwise parameters correlation (av erage o v er the 1000 replicates) with increasing sample sizes and different mo dels ( M 0 , M 1 and M 2 ). 20 n M 0 M 1 M 2 50 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.263 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.101 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.807 100 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.239 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.293 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.758 200 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.338 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.393 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.78 500 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.382 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.414 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.773 1000 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.376 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.464 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.8 1500 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.423 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.483 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.81 2000 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.353 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.493 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.809 3000 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.374 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.434 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −2 0 2 4 6 8 0.0 0.2 0.4 0.6 0.8 1.0 β 0 π corr= 0.801 Figure 4: Scenario (iii): scatterplot of π versus β 0 with increasing sample sizes and differen t mo dels ( M 0 , M 1 and M 2 ). T o v erify the predictiv e p erformance w e considered relativ e measures of sp ecificit y and 21 sensitivit y (F a w cett, 2006) build as the ratio of the same measures for M 2 (n umerator) and for M 1 (denominator) resp ectiv ely . In Figure 5 the obtained v alues are rep orted versus sample sizes. Remark that M 2 rapidly reac hes the same lev el of p erformance as M 1 with increasing sample size. ● ● ● ● ● ● ● ● 0 500 1000 1500 2000 2500 3000 0.80 0.85 0.90 0.95 1.00 1.05 1.10 sample size ● relative specificity relative sensitivity Figure 5: Scenario (iii): relative sp ecificity and sensitivit y computed as ratios b etw een M 2 and M 1 sp ecificit y and sensitivit y measures with increasing sample sizes. Dashed tra jectories are rep orted to sho w the patterns. 6 Conclusions In this w ork, w e presented a Bay esian pro cedure to estimate the parameters of logistic regressions for presence-only data. The approach w e prop osed is based on a t w o levels sc heme where a generating probabilit y law is combined with a case-control design ad- justed for presence-only data. The new formalization allows to consider rigorously all the mathematical details of the mo del as for instance the appro ximation of the ratio (15) that represen ts the crucial point when mo deling presence-only data in the finite p opulation setting. W e w ant to p oin t out that our formalization is substan tially differen t from the w ork by W ard et al. (2009), although we end with the same statistical mo del. W e con- cen trated on the case of the linear logistic regression b ecause we w ere a ware that some care is necessary to handle the identifiabilit y issues present in the mo del. The comparative simulation study considered three scenarios with differen t levels of com- plexit y across increasing sample sizes. W e presented detailed results with resp ect to the most difficult case where the contribution of an informative cov ariate w as mixed with a constan t effect and a white Gaussian noise. In term of p oin t estimation, the estimates based on our mo del w ere comparable to those obtained under the presence-only data b enc hmark in which the empirical p opulation prev alence was assumed to b e known. On the other hand, this lack of information on the p opulation prev alence affected the effi- ciency of the estimates, that resulted smaller for our mo del than for the b enchmark. This difference was significant only when the sample size n w as smaller than 1000, i.e. when the n um b er of observed presences n p w as smaller than 200. F rom the predictive p oint 22 of view, our mo del p erformed as well as the b enc hmark already for sample sizes ab out n = 200 , i.e. for a n umber of observed presences at least n p = 40 . Also the pairwise correlation b et w een β 0 and π , that represents an imp ortant issue as p ointed by W ard et al. (2009), b ecame negligible with increasing sample sizes. F rom the computational p oint of view, the pro cedure were carried out through a MCMC sc heme with data augmentation and implemented in F ortran co des. F uture work will in v estigate the p ossibility of adding dep endence structures among the p opulation units in to the mo del as, for instance, through the use of regression functions with structured random effects. 23 App endix Prop osition 3. Under the assumption that, giv en Y , the inclusion into the sample ( C = 1 ) is indep enden t from the co v ariates X , it results P r ( Y = 0 | C = 1 , x ) P r ( C = 1 | x ) = 1 − π ∗ ( x ) 1 + π ∗ ( x ) ρ 0 and P r ( Y = 1 | C = 1 , x ) P r ( C = 1 | x ) = 2 π ∗ ( x ) 1 + π ∗ ( x ) ρ 1 . Pro of. In general w e ha v e that P r ( Y | C = 1 , x ) = P r ( C = 1 | Y , x ) P r ( Y | x ) P r ( C = 1 | x ) (31) F rom the conditional indep endence b etw een C = 1 and X given Y , the (31) b ecomes P r ( Y | C = 1 , x ) = P r ( C = 1 | Y ) P r ( Y | x ) P r ( C = 1 | x ) . Recalling that P r ( Y = 1 | x ) = 2 π ∗ ( x ) 1+ π ∗ ( x ) and the definitions of ρ 0 = P r ( C = 1 | Y = 0) and ρ 1 = P r ( C = 1 | Y = 1) the pro ofs for Y = 0 and Y = 1 can b e derive by simple algebra. 24 n Model β 0 β 1 π 50 M 0 0.40 (-0.31 ; 1.45) 1.56 (1.08 ; 2.72) 0.20 (0.18 ; 0.25) M 1 2.19 (0.68 ; 3.57) 2.19 (1.37 ; 3.74) 0.23 (0.18 ; 0.27) M 2 1.03 (-2.51 ; 3.35) 2.00 (1.07 ; 3.44) 0.19 (0.12 ; 0.26) 100 M 0 0.31 (-0.18 ; 0.88) 1.23 (0.99 ; 1.61) 0.21 (0.19 ; 0.25) M 1 1.24 (0.29 ; 2.36) 1.55 (1.12 ; 2.22) 0.23 (0.19 ; 0.26) M 2 1.22 (-0.40 ; 2.69) 1.50 (1.07 ; 2.22) 0.16 (0.16 ; 0.27) 200 M 0 0.11 (-0.20 ; 0.46) 1.08 (0.95 ; 1.28) 0.22 (0.19 ; 0.24) M 1 0.46 (0.00 ; 1.27) 1.23 (0.99 ; 1.56) 0.22 (0.20 ; 0.24) M 2 0.48 (-0.24 ; 1.55) 1.23 (0.96 ; 1.59) 0.22 (0.19 ; 0.25) 500 M 0 0.06 (-0.10 ; 0.25) 1.02 (0.94 ; 1.12) 0.22 (0.20 ; 0.23) M 1 0.17 (-0.09 ; 0.47) 1.04 (0.92 ; 1.19) 0.22 (0.20 ; 0.23) M 2 0.14 (-0.26 ; 0.61) 1.03 (0.89 ; 1.20) 0.22 (0.19 ; 0.24) 1000 M 0 0.04 (-0.05 ; 0.17) 1.01 (0.95 ; 1.07) 0.22 (0.21 ; 0.22) M 1 0.04 (-0.13 ; 0.26) 0.99 (0.91 ; 1.08) 0.21 (0.21 ; 0.22) M 2 0.03 (-0.22 ; 0.34) 0.98 (0.90 ; 1.09) 0.21 (0.20 ; 0.23) 1500 M 0 0.05 (-0.04 ; 0.15) 0.99 (0.95 ; 1.04) 0.21 (0.21 ; 0.22) M 1 0.01 (-0.12 ; 0.18) 0.97 (0.91 ; 1.05) 0.21 (0.21 ; 0.22) M 2 0.00 (-0.24 ; 0.23) 0.97 (0.90 ; 1.05) 0.21 (0.20 ; 0.22) 2000 M 0 0.03 (-0.04 ; 0.10) 0.99 (0.95 ; 1.03) 0.21 (0.21 ; 0.22) M 1 0.00 (-0.12 ; 0.14) 0.97 (0.92 ; 1.03) 0.21 (0.21 ; 0.22) M 2 -0.02 (-0.22 ; 0.14) 0.96 (0.90 ; 1.02) 0.21 (0.20 ; 0.22) 3000 M 0 0.03 (-0.02 ; 0.10) 0.98 (0.96 ; 1.02) 0.21 (0.21 ; 0.22) M 1 0.00 (-0.10 ; 0.09) 0.96 (0.92 ; 1.00) 0.21 (0.21 ; 0.22) M 2 -0.03 (-0.18 ; 0.11) 0.95 (0.91 ; 1.00) 0.21 (0.20 ; 0.22) T able 5: Scenario (i): p oint estimates of regression parameters and prev alence computed as medians o v er 1000 replicates with increasing sample sizes and different mo dels ( M 0 , M 1 and M 2 ). In paren thesis distributions quartiles are rep orted. 25 n Model β 0 β 1 π 50 M 0 0.42 (-0.33 ; 1.42) 1.34 (0.94 ; 2.13) 0.23 (0.18 ; 0.28) M 1 2.12 (0.78 ; 3.39) 1.95 (1.25 ; 2.96) 0.24 (0.20 ; 0.28) M 2 1.26 (-2.99 ; 3.33) 1.75 (0.95 ; 2.79) 0.20 (0.13 ; 0.28) 100 M 0 0.19 (-0.20 ; 0.73) 1.07 (0.88 ; 1.35) 0.23 (0.19 ; 0.25) M 1 1.13 (0.36 ; 2.40) 1.39 (1.00 ; 1.96) 0.23 (0.21 ; 0.26) M 2 1.03 (-0.40 ; 2.65) 1.34 (0.96 ; 1.96) 0.22 (0.17 ; 0.28) 200 M 0 0.13 (-0.18 ; 0.45) 0.97 (0.84 ; 1.12) 0.23 (0.20 ; 0.25) M 1 0.48 (0.02 ; 1.17) 1.08 (0.88 ; 1.36) 0.23 (0.21 ; 0.25) M 2 0.48 (-0.38 ; 1.56) 1.07 (0.83 ; 1.41) 0.23 (0.18 ; 0.27) 500 M 0 0.09 (-0.07 ; 0.27) 0.92 (0.85 ; 1.00) 0.22 (0.21 ; 0.24) M 1 0.22 (-0.04 ; 0.53) 0.95 (0.84 ; 1.09) 0.23 (0.21 ; 0.24) M 2 0.23 (-0.24 ; 0.69) 0.95 (0.81 ; 1.09) 0.22 (0.20 ; 0.25) 1000 M 0 0.08 (-0.02 ; 0.20) 0.90 (0.86 ; 0.95) 0.22 (0.21 ; 0.23) M 1 0.09 (-0.07 ; 0.31) 0.90 (0.83 ; 0.99) 0.22 (0.21 ; 0.23) M 2 0.08 (-0.19 ; 0.38) 0.89 (0.82 ; 1.00) 0.22 (0.21 ; 0.24) 1500 M 0 0.08 (0.00 ; 0.18) 0.90 (0.86 ; 0.94) 0.22 (0.22 ; 0.23) M 1 0.08 (-0.05 ; 0.23) 0.89 (0.84 ; 0.96) 0.22 (0.22 ; 0.23) M 2 0.05 (-0.17 ; 0.30) 0.89 (0.82 ; 0.96) 0.22 (0.21 ; 0.23) 2000 M 0 0.07 (0.00 ; 0.15) 0.89 (0.86 ; 0.92) 0.22 (0.22 ; 0.23) M 1 0.06 (-0.06 ; 0.20) 0.89 (0.84 ; 0.95) 0.22 (0.22 ; 0.23) M 2 0.02 (-0.17 ; 0.24) 0.88 (0.82 ; 0.94) 0.22 (0.21 ; 0.23) 3000 M 0 0.07 (0.01 ; 0.13) 0.89 (0.87 ; 0.91) 0.22 (0.22 ; 0.23) M 1 0.04 (-0.04 ; 0.15) 0.88 (0.84 ; 0.92) 0.22 (0.22 ; 0.23) M 2 0.02 (-0.13 ; 0.17) 0.87 (0.83 ; 0.92) 0.22 (0.21 ; 0.23) T able 6: Scenario (ii): p oin t estimates of regression parameters and prev alence computed as medians o v er 1000 replicates with increasing sample sizes and different mo dels ( M 0 , M 1 and M 2 ). In paren thesis distributions quartiles are rep orted. 26 References Araùjo, M. and Williams, P . (2000). Selecting areas for sp ecies p ersistence using occur- rence data. Biolo gic al Conservation , 96 , 331–345. Armenian, H. (2009). The Case-Contr ol Metho d: Design And Applic ations . Oxford Univ ersit y Press, New Y ork, USA. Breslo w, N. E. (2005). Handb o ok of Epidemiolo gy , chapter 6: Case-Con trol Studies, pages 287–319. Springer, New Y ork, USA. Breslo w, N. E. and Dey , N. E. (1980). Statistic al Metho ds In Canc er R ese ar ch, V olume 1 - The analysis of c ase-c ontr ol studies . WHO In ternational Agency for Researc h on Cancer, Ly on, F rance. Chakrab ort y , A., Gelfand, A. E., Wilson, A. M., Latimer, A. M., and Silander, J. A. (2011). Poin t pattern mo delling for degraded presence-only data ov er large regions. Journal of the R oyal Statistic al So ciety: Series C (Applie d Statistics) , 5 , 757–776. Di Lorenzo, B., F arcomeni, A., and Golini, N. (2011). A Bay esian mo del for presence-only semicon tin uous data with application to prediction of abundance of Taxus Baccata in t w o Italian regions. Journal of A gricultur al, Biolo gic al and Envir onmental Statistics , 16 (3), 339–356. Divino, F., Golini, N., Jona Lasinio, G., and P ettinen, A. (2011). Data augmentation ap- proac h in bay esian mo delling of presence-only data. Pr o c e dia Envir onmental Scienc es , 7 , 38–43. Dorazio, R. M. (2012). Predicting the geographic distribution of a sp ecies from presence- only data sub ject to detection errors. Biometrics , 68 , 1303–1312. Elith, J. and Leathwic k, J. R. (2009). Sp ecies distribution mo dels: ecological explana- tion and prediction across space and time. Annual R eview of Ec olo gy, Evolution and Systematics , 40 , 677–697. Elith, J., Graham, C. H ., Anderson, R. P ., Dudik, M., F errier, S., Guisan, A., Hijmans, R. J., Huettmann, F., Leath wic k, J. R., Lehmann, A., Li, J., Lohmann, L. G., Loiselle, B. A., Manion, G., Moritz, C., Nak am ura, M., Nak azaw a, Y., Overton, J. M., P eterson, A. T., Phillips, S. J., Ric hardson, K. S., Scac hetti-P ereira, R., Sc hapire, R. E., Sob eron, J., Williams, S., Wisz, M. S., and Zimmermann, N. E. (2006). Nov el metho ds impro ve prediction of sp ecies’ distribution from o ccurence data. Ec o gr aphy , 29 , 129–151. Elith, J., Phillips, S. J., Hastie, T., Dudík, M., Chee, Y. E., and Y ates, C. J. (2011). A statistical explanation of MaxEnt for ecologists. Diversity and Distributions , 17 , 43–57. F aw cett, T. (2006). An introduction to ROC analysis. Pattern R e c o gnition L etter , 27 , 861–874. 27 F ranklin, J. (2010). Mapping Sp e cies Distributions: Sp atial Infer enc e And Pr e diction . Cam bridge Univ ersity Press, Cambridge, UK. Ja ynes, E. T. (1957). Information theory and statistical mechanics. The Physic al R eview , 106 (4), 620–630. Keating, K. A. and Cherry , S. (2004). Use and in terpretation of logistic regression in habitat-selection studies. Journal of Wild life Management , 68 , 774–789. Lancaster, T. and Im b ens, G. (1996). Case-control studies with con taminated con trols. Journal of Ec onometrics , 71 , 145–160. Little, R. J. A. and Rubin, D. B. (1987). Statistic al Analysis With Missing Data . John Wiley & Sons, New Y ork, USA. Liu, J. S. (2008). Monte Carlo Str ate gies In Scientific Computing . Springer, New Y ork, USA. Liu, S. Y. and W u, Y. N. (1999). Parameter expansion for data augmentation. Journal of A meric an Statistic al Asso ciation , 94 , 1264–1274. P earce, J. L. and Boyce, M. S. (2006). Mo delling distribution and abundance with presence-only data. Journal of Applie d Ec olo gy , 43 , 405–412. Phillips, S. J., Anderson, R. P ., and Sc hapire, R. E. (2006). Maxim um entrop y mo deling of sp ecies geographic distributions. Ec olo gic al Mo del ling , 190 , 231–259. Rob ert, C. P . and Casella, G. (2004). Monte Carlo Statistic al Meto ds . Springer, New Y ork, USA. Rubin, D. B. (1976). Inference and missing data. Biometrika , 63 (3), 581–592. Särndal, C. E. (1978). Design-based and mo del-based inference in survey sampling. Sc an- dinavian Journal of Statistics , 5 , 27–52. T anner, M. (1996). T o ols for Statistic al Infer enc e: Observe d Data And Data A ugmenta- tion . Springer, New Y ork, USA. T anner, M. and W ong, W. (1987). The calculation of p osterior distribution b y data augmen tation. Journal of A meric an Statistic al Asso ciation , 82 , 528–550. W ard, G., Hastie, T., Barry , S., Elith, J., and Leathwic k, A. (2009). Presence-only data and the EM algorithm. Biometrics , 65 , 554–563. W arton, D. I. and Shepherd, L. (2010). P oisson p oint p orcess mo dels solve the “pseudo- absence problem” for presence-only data in ecology . Annals of Applie d Statistics , 4 (3), 1383–1402. 28 W o o dward, M. (2005). Epidemiolo gy: Study Design A nd Data A nalysis . Chapman & Hall, New Y ork, USA. Zaniewski, A. E., Lehmann, A., and Ov erton, J. M. (2002). Prediction sp ecies spatial dis- tributions using presence-only data: a case study of nativ e New Zeland ferns. Ec olo gic al Mo del ling , 157 , 261–280. 29