估計空間群聚購物行為對方便品購買之影響

Abstract

在本研究中，探討顧客在空間上之購物行為，結合了地理和行銷的概念，屬於地理行銷（Geomarketing）的問題。我們想要探討（1）基於空間便利性，購買地點越群聚者是否會越常購買方便品？（2）基於時間便利性，購買地點越不群聚者是否一次會購買越多方便品？探討以上假說時，該研究會出現自我選擇偏誤（self-selection bias）問題。因為實驗組和控制組 - 方便品之空間群聚購物行為者和非空間群聚購物行為者，這兩組人在本質上是不同的。因此，本研究透過傾向評分匹配（Propensity Score Matching, PSM）方法，結合顧客基本資料來平衡兩組，以得到更精確的影響。在模擬研究中，某些情況下，線性機率模型（Linear Probability Model, LPM）在執行PSM時具有與邏輯模型相似的結果，促使我們運用LPM於實證研究中。另外，也因邏輯模型無法處理連續型依變數，因此，在實證研究中，我們更使用連續變數，即顧客之購買地點群聚程度作為處理，並將LPM視為在此情況下與邏輯模型互補的方法。

最後我們發現，整體來說，當顧客具有空間群聚購物行為，會基於空間便利越常去買方便品，同時也會買越多。但仍有一些具有空間群聚行為的顧客較不常購買，因此，我們便推薦這些顧客在其個人化購物熱區，基於空間便利性更頻繁地購買方便品。

Geomarketing is a combination of Geography and Marketing, associated with our study of customer spatial shopping behavior. However, the self-selection bias problem will occur in Geomarketing. In our empirical study, the treatment and control group– spatially clustered and spatially non-clustered shopping behavior on convenience goods differ from each other essentially. Fortunately, we can use the characteristics of customers to balance two groups by Propensity Score Matching (PSM). In our simulation study, we show Linear Probability Model (LPM) in PSM has similar performance to the logit model in some conditions. Besides, we use continuous variable, clustering degree to be the treatment, regarding LPM as a complementary approach to the logit model in such case.

Finally, we find that customers with spatially clustered behavior will buy more often and do not pay significantly more on convenience goods, so they have emergent and spatially clustered shopping behavior. For bank, the strategy is to promote them to purchase all convenience goods they need at one time, and they do not need to shop whenever some products come to their minds. Besides, the bank can further promote them to shop in their personalized shopping hotspot based on their spatial convenience.