Our research provides a unique approach to evaluate the online consumers purchasing behavior patterns using the session level panel data by investigating the impacts of return purchase, web browsing characteristics and price strategy on consumers’ purchasing willingness. Through establishing an empirical research model, we verify the propositions proposed based on extant studies. We find that return purchase behavior would decrease consumers’ purchase willingness. The length of website visit duration is positively related with purchase willingness while page view is not. Consumers are likely to be influenced by price strategy in the current session when shopping online, while less influenced by previous experience. In addition, we find that consumers willing to pay more money online are quite purposeful. The quantity as well as types of products they buy in one session is relatively fewer. Also they usually visit target website directly without referring to other domain.
The detailed design of this research.
In particular, this paper mainly addresses the following research questions:
(1) How would return purchase influence consumers’ purchase willingness? Would the purchase willingness increase or decrease with more repeated purchases happen in one website?
(2) How would web browsing behaviors influence consumers’ purchase willingness? Would the purchase willingness increase or decrease with longer visit duration and more page views to the website?
(3) How would price strategy influence consumers’ purchase willingness? Would the purchase willingness increase or decrease when websites conduct certain price strategies?
Based on theoretical discussion, in this section we would establish and specify the purchase willingness model aiming at solving the research questions of this study. To measure the contribution of the different inputs we take into account, we form the empirical model on the basis of individual-specific effects model.
lpurchaseit= αi+ β1returnit+ β2durationit + β3pageit + β4strategyit + β5fstrategyit + β6refit + β7qtyit + β8typeit + εit
In this equation, lpurchaseitis the output measuring consumers’ purchase willingness in one session. The variable lpurchaseit shows a consumer’s tolerance to the highest price they are willing to pay online. The consumer is more likely to spend more money and buy high value products online if the person’s purchase willingness is high. Since we could not further conduct survey to examine the intrinsic feelings of the participants, this variable is appropriate to measure purchase willingness from the objective approach. In the right side of the equation, αi are random individual-specific effects that capture unobserved heterogeneity. Furthermore, we would estimate this purchase willingness equation by fixed effects estimation and random effects estimation. In fixed effects model, αi represents unobserved random variable that is potentially correlated with the observed regressors. In random effects model, αi represents the unobservable individual effect that is distributed independently of the regressors. Besides, εit represent idiosyncratic error terms which are independent and identically distributed (i.i.d.) over i and t.
This model incorporates eight explanatory variables, namely return, duration, page, strategy, fstrategy, ref, qtyand type, and a set of β are coefficients for these variables. Among them, return represents whether the transaction happens in each session is return purchase, which is a dummy variable with “1” indicating return purchase and “0” indicating first time purchase in this website. Besides, duration and page represent visit duration at a site and the number of pages viewed in each session respectively. In addition, strategy and fstrategy represent the price strategy in the current session and the price strategy conducted by the website of the last purchase respectively, and we can notice fstrategy is the lagged variable of the strategy variable. It demonstrates that when people are shopping online, their purchase willingness is determined by both current and previous price strategy. They are also dummy variables with “1” indicating websites are offering discount to consumers and “0” indicating websites are charging additional fee along with product net price. What’s more, ref is a dummy variable with “1” indicating people refer to other domain in this session and “0” indicating no reference behavior occurs in this session, and qtyand type represent the total quantity of products and the number of types of products purchased in each session respectively.
In this study, we use the ComScore 2004 disaggregate dataset which captures detailed browsing and buying behavior for over 20000 internet users across the United States.
We focus on the 12-month period from January 2004 to December 2004, and we further aggregate the data to the session level. Among the 52,028 households joining in this program, 24,831 of them once purchased online in 2004. During this period, 114,393 sessions are in record and all these sessions contain actual purchase behavior. In this study, we would perform further analysis based on these 114,393 session records with purchasing activities.
Summary Statistics of Variables (Overall)
After the data preparation is ready, we further estimate the established purchase willingness equation by using within estimation (fixed effects estimation) and random effects estimation, and compare which one is more appropriate and provides consistent estimates. We would list the results of both two estimations in the following table for further analysis and reference.
We further perform a Hausman specification test to evaluate the significance of fixed effects estimator versus random effects estimator. The result of hausman test lists the coefficients estimated by the two models. For the coefficients of variable return, a test of random effects estimation against fixed effects estimation yields t=12.003 (0.039729/0.00331), a highly statistically significant difference. Both FE and RE are consistent if the regressors are uncorrelated. However, if regressors are correlated, FE is consistent while RE is inconsistent. Here the large difference on return between FE and RE is taken as evidence that the regressors could be correlated. In addition, the overall statistic χ2(8) is significant (p=0.0000). This leads to strong rejection of the null hypothesis that RE model parameters are consistent and efficient. Consequently, we would base the further analysis on the output of fixed effects estimation which is more appropriate.
This study attempts to explore factors influencing online consumers purchasing behavior patterns by establishing a purchase willingness model analyzing observed online clickstream data. While previous studies focused on investigating the internet clickstream data collected for a single site, this study captures consumers’ shopping behavior across multiple websites which is more comprehensive and close to the reality. Based on the empirical results, return purchase would decrease consumers’ purchase willingness. If people come to the online store for the second time or more, they are less likely to spend as much as money as they did when they first came. It indicates that there could be such initial novelty effects that slowly wear away for online consumers. Therefore consumer retention is actually more difficult to achieve online especially for those who regard shopping as a chance for socializing and having fun. Another proof to this point is consumers are more intended to be influenced by price strategy in current session rather than previous session. This reveals online consumers are profit-oriented and less likely to stick with only one or two websites they are familiar with. Motivated by strong curiosity, they are always willing to try something new. Some shoppers might purchase many times in one website just because it provides price benefit, and once the website stops offering such discount consumers would switch to other websites, showing low consumer retention again. Besides, we prove duration is positively related to purchase willingness, indicating that consumers are enjoying such experience and oftentimes losing the sense of physical time. Moreover, online consumers viewing fewer pages are willing to spending more money, indicating they do focus on the content and take more time finding useful information in each page. The findings on the impacts of domain reference, product quantity and product type show that consumers with clear purchasing purpose possess higher purchase willingness. They purchase fewer products online in one session but the mean price they pay for each unit is high.
Future studies in this field could further investigate this issue by taking account into more characteristics of the multiple sites and products. For example, we could select several most popular product categories and choose the representative online stores to take a closer look at how these factors influencing consumer purchase willingness. Furthermore, we could compare the result of one industry with another to see how industries and the companies within differ from each other.
For website managers who wish to expand their market share, they should focus on attracting new consumers with price advantages. Other effective approaches include encouraging visitors to participate in various online activities, improve the quality of content quality and changing the website design more frequently to keep the website fresh and appealing to consumers.