پیش بینی و تبیین رفتار حمایت به سمت فروشگاه های وب و سنتی با استفاده از شبکه های عصبی: تجزیه و تحلیل مقایسه با رگرسیون لجستیک
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|24715||2006||18 صفحه PDF||سفارش دهید||9420 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Decision Support Systems, Volume 41, Issue 2, January 2006, Pages 514–531
Web stores, where buyers place orders over the Internet, have emerged to become a prevalent sales channel. In this research, we developed neural network models, which are known for their capability of modeling noncompensatory decision processes, to predict and explain consumer choice between web and traditional stores. We conducted an empirical survey for the study. Specifically, in the survey, the purchases of six distinct products from web stores were contrasted with the corresponding purchases from traditional stores. The respondents' perceived attribute performance was then used to predict the customers' channel choice between web and traditional stores. We have provided statistical evidence that neural networks significantly outperform logistic regression models for most of the surveyed products in terms of the predicting power. To gain more insights from the models, we have identified the factors that have significant impact on customers' channel attitude through sensitivity analyses on the neural networks. The results indicate that the influential factors are different across product categories. The findings of the study offer a number of implications for channel management.
The Internet is changing the way firms market and distribute their products to customers. Despite the fact that Internet bubble in 2002 was accompanied by the shutdown of many Internet companies, sales over the Internet have continued to increase. According to Forrester Research , online sales in the United States grew 51% to approximately US$26 billion just in the third quarter of 2003. Seemingly, web stores, where buyers place orders over the Internet, have emerged to become a prevalent sales channel. While more and more companies are engaging in online sales, there are speculations of an uncertain future of e-commerce due to the fact that the total amount of online sales is still a small portion of total retail sales. According to the U.S. Census Bureau , online sales accounted for only 1.6% of all retail sales in 2003. Will web stores prevail in future? Apparently, the success of a web store as a viable sales channel is dependent upon whether it helps to attain a significant amount of potential customers who are willing to make purchases online. Therefore, understanding consumers' attitude toward web stores appears crucial in the business-to-consumer (B2C) e-business context. The questions are: What are the predictors of consumers' online buying behavior? Are we able to accurately predict and explain consumers' channel choice between web and traditional stores? As indicated by Chiang et al. , the answers to the questions provide significant implications for firms who want to expand their market potential by tapping into customer segments that otherwise would not buy, or for suppliers who are strategically contemplating multi-channel distributions. Although there are some recent papers (e.g., Refs.  and ) that provide insights into customers' channel choice through analytical models and game theories, most studies seeking to address the above questions are based on empirical surveys and statistical analyses. For example, Liang and Huang  tried to explain the acceptance of online buying using consumer perceptions of transaction-costs associated with shopping, uncertainty and asset. The authors provided evidence that, in general, customers prefer traditional markets to the web stores and different products have different customer acceptance levels on the electronic market. Szymanski and Hise  measured “satisfaction” with the Internet-shopping experience in a study of antecedents of e-satisfaction. They found that greater satisfaction with online shopping is positively correlated with consumer perceptions of the convenience, product offerings, product information, site design and financial security of web stores relative to traditional stores. Degeratu et al.  studied the decisions of individuals to use Peapod online grocery shopping. They gathered a sample of Peapod online buyers and a matching sample of individuals who did their grocery shopping in traditional supermarkets. As part of their broader study of brand preferences, their random utility model specified an indirect utility function for online versus offline shopping that depended only on the income of individuals. Bellman et al.  analyzed the responses of over 8000 participants in the Wharton Virtual Test Market who completed an initial survey about online buying and attitudes. Their logistic regression model indicated that online experience (i.e., web browsing) was the dominant predictor of whether or not a respondent had ever bought anything online. Kwak et al.  surveyed chatroom participants via email to discover whether these consumers had bought any of nine products online. They showed that four broad independent constructs (attitudes toward the Internet, experience with the Internet, demographics, and personality type) could explain Internet purchases of those products with logistic regressions. All of the above empirical studies are forms of what Urban and Hauser  called “preference regressions” and they all share the same a priori assumption that the process of consumers' channel evaluation is linear compensatory. Specifically, those models assume that any shortfall in one channel attribute (e.g., immediate possession of a product) can be compensated by enhancements of other channel attributes (e.g., price). Although linear compensatory models, which can be easily estimated by statistical methods (such as analysis of variance procedures, logistic regression, and discriminant analysis), are widely used to predict consumer behavior for their ability to imitate consumer choice processes, challenges regarding their reliability have been levied by many research studies. It has been demonstrated that consumers might judge alternatives based on only one or a few attributes, and therefore the process of evaluation might not always be compensatory  and . For instance, in the case of channel choice, the consumers' concern may just be immediate possession of a product. This concern may not be compensated by the enhancement of other channel attributes, such as price (consumers do not mind paying more to possess a product immediately from another channel). Johnson et al.  suggested that compensatory statistical models may not be able to capture noncompensatory decision rules and, consequently, may be unreliable. To the best of our knowledge, there are no research studies that have used noncompensatory models to explain consumers' channel choice between traditional and web stores. Against this backdrop, this paper is motivated by the intention of making a contribution to this important line of inquiry. Specifically, we develop neural network models, which are known for their capability of modeling noncompensatory decision processes, to address the following research questions: Do noncompensatory choice models using neural networks perform better than logit choice models in predicting consumers' channel choice between web and traditional stores? If so, based on the noncompensatory choice models, what are the main predictors of customers' online buying behavior? Overview of neural networks for noncompensatory decision processes Artificial neural networks are computer models used to emulate the human pattern recognition function through a similar parallel processing structure of multiple inputs. They learn the intrinsic nature of a pattern or process from sample data. A neural network consists of a set of fundamental processing elements (called nodes or neurons) that are distributed in a few hierarchic layers. Most neural networks contain at least three normal types of layers–input, hidden and output. The layer of input normally receives the data either from input files or directly from electronic sensors in real-time applications. The output layer generates information or conclusions. Between these two layers can be a number of hidden layers. In most networks, after each neuron in a hidden layer receives inputs from all of the neurons in a layer above it, typically an input layer, the values are added through applied weights and converted to an output value by a node activation function. Then, the result is passed to all of the neurons in the layer below it, providing a feed-forward path to the output layer. The weights of connections between two neurons in two adjacent layers are adjusted through an iterative training process where training samples are presented to the network. They are used to store knowledge and make it available for future use. Characterized by the pattern of connections between neurons, the method of determining weights on connections, and a node activation function, a neural network is designed to capture causal relationships among dependent and independent variables in a given sample data set. Unlike parametric models used in statistical techniques, neural networks do not require any restrictive a priori assumptions about the relationship among independent and dependent variables. In addition, they are adaptive and can respond to structural changes in the data generation process in ways that parametric models cannot. Neural networks have been heavily used to model business problems in support of finance and marketing decision-making  and . In most of those applications, neural networks outperformed traditional compensatory models such as discriminant and regression analysis ,  and . In this study, we derived similar results in a different context. Based on the data that we collected through an empirical survey, we found that, in general, the noncompensatory neural network models outperform the compensatory logit choice models in terms of accuracy in predicting consumers' channel choice between web and traditional stores. The remainder of this paper is organized as follows. In the next section, we outline the empirical survey procedures and present the demographic data of the survey respondents. Then, we explain channel attributes and product categories used in the survey and report preliminary survey outcome. The logit choice models that are used to establish a performance benchmark are then introduced in the section that goes after, and it is followed by the section that presents the neural network models of consumer channel choice. Later, we report the results of our investigation and discuss some managerial implications. The paper is concluded with a summary of the findings in this study.
نتیجه گیری انگلیسی
In this paper, we developed neural networks and logistic regression models to predict and explain consumers' choice between web and traditional stores. In order to identify new predictors of customers' online buying behavior, we conducted an empirical survey for the study. Specifically, in the survey, purchases from web stores were contrasted with purchases from traditional stores for six distinct product categories. The respondents' perceived attribute performance was then used to predict customers' channel choice between web and traditional stores. We have provided statistical evidence that neural networks significantly outperform logistic regression models for most of the selected products in terms of the predictive power. To gain more insights and implications from the models, we have identified the factors that have significant impact on customers' channel choice through sensitivity analyses in the neural networks for each of the surveyed products. The results indicate that the influential factors are different across product categories. The findings of the study help us understand the decision support needs in online marketing and customer relationship management. For example, the improvement on some store characteristics may have little effect on consumer patronage for some products, and therefore, should not be high priorities for managerial actions. On the other hand, some shopping behaviors are strongly influenced by other variables that deserve more managerial attention and improvement. As indicated by , important decision support issues need to be tackled once a marketing channel decision has been made. Understanding what factors have the most significant impact on customers' channel choice appears to be very critical in providing a decision support framework for shopping store management. Web stores empower consumers with the ability to make informed decisions. However, the advantages of web stores may be dampened by their inherent limitations and consumers' fear of the web. In addition to improving web stores' service quality, educating the public on basic skills of using the web is also important. Traditional stores survived and will continue to survive. Findings in this study suggest that some types of products are more favorable for shopping online than others, and online consumers may value channel attributes differently from traditional store consumers for the same product categories. Therefore, in order to gain more competitive advantages, stores should focus on improving store attributes that are perceived important by consumers of the corresponding products in the corresponding channel. On the other hand, being aware of the strength of the opposite channel may also help managers better position themselves and make strategic decisions for their own stores. We are not aware of any extant research studies using non-compensatory models to predict and explain consumers' channel choice between traditional and web stores. While we believe that the neural network models developed in this paper and the implications of our results are important contributions to the related literature, there is still scope for further work in this area. For example, users' personal traits, such as Internet experience, computer skill, and cognitive style, may be used for prediction of user online behavior. In this paper, we did not perform further demographical analysis due to the limitation on the data applicability. Clearly, studies seeking to analyze channel choice based on demographic categories would be valuable to extend this research.