Collective Attention and the Dynamics of Group Deals

We present a study of the group purchasing behavior of daily deals in Groupon and LivingSocial and introduce a predictive dynamic model of collective attention for group buying behavior. In our model, the aggregate number of purchases at a given time comprises two types of processes: random discovery and social propagation. We find that these processes are very clearly separated by an inflection point. Using large data sets from both Groupon and LivingSocial we show how the model is able to predict the success of group deals as a function of time. We find that Groupon deals are easier to predict accurately earlier in the deal lifecycle than LivingSocial deals due to the final number of deal purchases saturating quicker. One possible explanation for this is that the incentive to socially propagate a deal is based on an individual threshold in LivingSocial, whereas in Groupon it is based on a collective threshold, which is reached very early. Furthermore, the personal benefit of propagating a deal is also greater in LivingSocial.

💡 Research Summary

The paper investigates the purchasing dynamics of daily group‑deal platforms—Groupon and LivingSocial—by analyzing large‑scale transaction data and proposing a stochastic model that separates two fundamental processes: random discovery and social propagation. Random discovery captures users who encounter a deal through the website, mobile app, or email, while social propagation accounts for purchases triggered by friends sharing the deal. The authors argue that these two processes are demarcated by a clear inflection point in the cumulative purchase curve. For Groupon this inflection coincides with the “tipping point” (the minimum number of purchases required for the deal to become valid), whereas LivingSocial lacks a formal tipping point but introduces a personal incentive: a user’s deal becomes free once three friends also purchase it.

Data collection covered 60 randomly selected U.S. cities over a two‑month period, yielding 4,376 Groupon deals, and a one‑month crawl of LivingSocial that produced over 900 deals. The authors filtered out anomalous deals that exhibited non‑monotonic purchase counts, leaving 3,876 Groupon deals for analysis. They first performed a multivariate linear regression of final purchase counts on deal attributes (tipping point, featured position, duration, price, discount, limited inventory, launch day, category, city). The regression revealed that tipping point and featured position are the most significant predictors, with tipping point showing a particularly strong effect.

Building on these observations, the authors formulate a dynamic model for the number of purchases N(t) at time t. In a discrete interval Δt the change ΔN(t) is expressed as:

ΔN(t) = α_t·Y_t + β_t·f(t, N(t))

where Y_t is a non‑negative random variable representing purchases from random discovery (modeled as a Poisson or geometric process), f(t, N(t)) is a function proportional to the current cumulative purchases that captures social propagation, and α_t, β_t are time‑varying weights. The model parameters are estimated via maximum likelihood on the observed purchase trajectories. Empirically, α_t dominates before the inflection point, while β_t rises sharply after the point, confirming the hypothesized shift from discovery‑driven to propagation‑driven growth.

Prediction experiments evaluate the model’s ability to forecast cumulative purchases at early stages (2 h, 6 h, 12 h after launch). Baselines include simple linear regression, ARIMA time‑series forecasting, and a generic log‑normal attention model derived from prior work on Digg and Twitter. The proposed model consistently outperforms baselines, achieving lower mean absolute error (MAE) and mean absolute percentage error (MAPE). For Groupon, the model reaches >85 % prediction accuracy within the first four hours, reflecting the rapid saturation of the tipping point (average 22 purchases, typically reached within a day). LivingSocial, by contrast, shows slower early growth due to the personal free‑deal incentive but catches up after about twelve hours, where its prediction accuracy matches that of Groupon.

The authors discuss practical implications. On Groupon, marketers should aim for low tipping thresholds and secure featured placement early to maximize random discovery and trigger the early inflection. On LivingSocial, emphasizing the “share‑to‑get‑free” incentive and possibly augmenting it with timed bonuses can accelerate the onset of social propagation. Moreover, the model can be embedded in real‑time dashboards to monitor deal performance, flag under‑performing offers, and enable dynamic budget reallocation or promotional adjustments.

In conclusion, the study provides a rigorous quantitative framework for understanding and forecasting group‑deal success, highlighting the distinct roles of random discovery and social contagion, and demonstrating how platform‑specific incentive structures shape the timing and magnitude of the inflection point. The work bridges gaps between social purchasing behavior literature and collective attention modeling, offering both theoretical insights and actionable tools for practitioners in the rapidly evolving daily‑deal market.