Consumer Behaviour For Purchasing Online Group Deals Marketing Essay

Published Date: 23 Mar 2015

Groupon, which is derived from the combination of "group coupon", is a newly rising form of ecommerce market. According to latest iResearch China online group shopping research report (2011), groupon website provides consumers with an extensive platform to purchase at a comparatively low cost. Because groupon is perceived as a form of electronic business rather than the specific website "Groupon" (groupon.com, which is the original and dominant company of this form of ecommerce), this paper prefers to use 'online group purchasing' to define this category of e-business, which can be helpful to differentiate from the specific website.

In the market of China, data (iResearch, 2011) indicated that 30.2% of total Chinese internet users, nearly 140 million consumers participated in online group purchasing in 2010. Also, the volume of trading amounted to 1.45 billion in 2010. Among the online group purchasing merchants, the website 'Lashou' led in the competition from most indicators. The tension was enhanced by participation from several social network services providers, such as sina Weibo, Renren, Kaixin, etc. Online group purchasing provides a platform for small or medium business to advertise on a low budget, which is the informational value of these coupons (Narasimhan, 1984). This business mode will probably become an influential one in the electronic market.

Groupon, coupon and deals can be seen as similar but slightly different promotion instruments of merchants. Essentially, online group purchasing is a category of coupon distributed by Internet. Online group purchasing websites act as an intermediary of business, they connect the merchants who are willing to provide deals and select reliable deals to post online. Initially, one deal would be offered in a specific city on a daily basis. As the commercial scale booming, the number of deals rose and available cities increased. Once the customers decide to take the advantage of deals, they need to pay online first, then they will receive a voucher code, after that, they redeem the voucher directly from the merchants. Additionally, the deals are only activated while the numbers of purchasing reach the pre-setting limit, for example, 10 consumers. The websites will proportionally charge agency fee as a medium.

The main difference between coupons and online group deals is payment timing. Traditionally, coupons are paid when consumers enjoy the services or goods, while online group deals have to pay upfront. This is also the difference between ecommerce and traditional offline business. Because of this difference, the analysis of redemption rate is thereby changing into analysing the purchasing rate of online deals. Due to convenience, the remaining essay will not distinguish these two definitions. That means, redemption rate of online group deals also means the purchasing rate. More importantly, problems of trust will be generated under such environment, which possibly influence the redemption rate (Brynjolfsson and Smith, 2000). Grabner-Kraeuter (2002) even suggests trust is the determinant of electronic business.

Researchers reveal that the main purpose of couponing is informing the existence of the merchants and attracting trial-repeat purchase (Narasimhan, 1984; Edelman et al. 2010). Similarly, online group deals share the same purpose and benefit. Casual empirical evidences that little famous merchants offer discount on Groupon support this argument. Moreover, Mulhern and Padgett (1995) provide empirical evidence that support one of the tasks of online group purchasing is advertising. They analyse the relationship between discount-priced products and regular-priced products. The statistically significant result shows consumers tend to spend more on the regular price goods when they are spending money on the promotion goods. Namely, by offering discount promotion, the merchants attract more consumers. That may be the reason why online group purchasing is so heated and attracts both consumers and retailers.

Previous studies mainly concentrate on coupons/deals rather than online group purchasing due to the novelty of e-deals. Prior research is productive in the motivation analysis of redeeming a coupon. Initially, scholars pay attention to the demographics (Webster, 1965; Blattberg et al. 1978; Narasimhan, 1984). Later, customer loyalty captures the researchers' attraction (Webster, 1965; Shoemaker and Tibrewala, 1985; Bawa and Shoemaker, 1987a). A strong and clear relationship between loyalty and redemption rate is constructed, while the demographics' effects are still mysterious. In the meanwhile, face value (Ward and Davis, 1978; Shoemaker and Tibrewala, 1985; Bawa and Shoemaker, 1987b; Bawa et al. 1997), underlying products (Bawa et al. 1997) and expire date (Inman and McAlister, 1994) are found that these features adhere the coupons will result in various possibilities when consumers redeem them.

Gaps in prior research

There are a variety of promotion methods not only online but also offline market. Due to the similar nature and essence of these promotion instruments, Blattberg and Neslin (1990) suggests all coupons and online group purchasing deals should be named as deal, and those who are incentive to seek promotions should be regarded as deal prone consumers. However, they believe different types of deals can generate different responses. In other words, consumer behaviour may be different and specified relating to the exact category of deals. Moreover, Mayhew and Winer (1992, cited in Lichtenstein et al. 1997) result that a household probably reacts positively to coupons, but is less willing to take other promotions instruments. Therefore, although online group purchasing deal is similar with coupons, it is necessary to study on this specific type separately.

Secondly, past articles on online group deals focused on the profitability of groupon rather than the customers' purchasing behaviour (Edelman et al. 2010; Gupta, 2012). That is a perspective from the operator. It will be necessary to examine the consumer behaviour in such field.

Thirdly, previous studies analysing redemption behaviour are mostly based on the unit of household. As the online group purchasing activity relies on technology knowledge and attracts more youngsters who are not married, this analysis may be better based on the unit of individual person. Additionally, the focus of this paper is Chinese market.

Therefore, this article aims to analyse the consumer purchasing behaviour of online group deals and provide a behaviour decisive factors model for online group shopping. Basically, the demographic characteristics of consumers will be examined. Moreover, this essay will define a new concept, channel loyalty, and test the relationship between this variable and online group deal proneness. Similar to brand/store loyalty, channel loyalty infers that consumers become loyal to a specific shopping channel, for example, online shopping, which indicates that consumer's preference level will remain as they keep shopping online increasingly.

In this paper, we argue that demographics may affect but not the decisive factors of consumer behaviour. Consumer behaviour is also influenced by the past purchasing experience, especially channel loyalty. Hypotheses will be assumed according to the past research and tested by statistics. Finally, the result supports the effect of channel loyalty but the results on demographics are mostly not significant. In the second part, past research on consumer behaviour and the distinct features of such new environment, ecommerce, will be demonstrated. In the third and fourth part, hypotheses and research method will be mentioned. After that, the fifth and sixth part will indicate empirical test results and relative discussion. In addition, a purchasing behaviour model will be derived in the sixth part. Managerial implication will be recommended in the seventh part. Lastly, the limitation of this paper and future studies' direction will be discussed.

2. Literature review

Studies focused on online group purchasing are limited. As an advanced form of coupons/deals, purchasing the group deals is essentially similar with redeeming a coupon, which just makes the payment in advance. Thus, to analyse online group purchase behaviour, review of prior research should concentrate on the redemption behaviour of coupons/deals. Meanwhile, ecommerce market features should be taken into consideration while discussing the distinct parts.

2.1 Redemption Behaviour

Prior studies on consumer behaviour towards coupons or deals focus on two main aspects. One is redemption behaviour of coupons or deals, these studies often analyse empirical data or construct models to demonstrate which categories of consumers are deal prone (Webster, 1965; Montgomery, 1971; McCanne, 1974; Cotton and Babb, 1978; Blattberg et al. 1978; Teel et al, 1980; Narasimhan, 1984; Shimp and Kavas, 1984; Shoemaker and Tibrewala, 1985; Bawa and Shoemaker, 1987a; Mittal, 1994). The second aspect is repeat purchasing (Shoemaker and Shoaf, 1977; Bawa and Shoemaker, 1987b; Neslin and Shoemaker, 1989; Graham, 1994; Taylor, 2001).

In terms of redemption behaviour, two streams can be isolated (Bawa et al. 1997). First, the main stream is from the aspect of consumer. For instance, analyse the consumer's demographics and other characteristics. Second, concentrate on the coupons or deals themselves, such as expiration date, attractiveness, face value analysis.

2.1.1 From the perspective of consumer

In terms of consumer's actions, according to Mittal's framework (1994, pp.533), redemption behaviour can be analysed by four categories of "individual-difference variables (IDVs)":

"Objective IDVs": basic demographics, such as income, educational level, family size, age, etc.

"Subjective IDVs": what consumers self-perceive they are, such as involvement degree of media, creativity, risky spirit, etc.

"Domain IDVs": factors influence their purchase behaviour, such as past purchase experience, sensitivity of price, etc.

"Cost or benefit perceptions": personal attitude towards the gain or loss of purchasing, such as consumer surplus, consumer satisfaction.

Under this framework, topics that based on objective IDVs and domain IDVs are most frequent targets in prior research. Basically, the review of previous consumer redemption behaviour studies will follow this framework.

2.1.1.1 Demographics

Firstly, a huge number of scholars pay attention on demographics. However, the result of this aspect remains controversial. In early studies, Webster (1965) firstly sets up a measurement of deal proneness and then run regressions to correlate the data. He finds that four demographic characteristics, income, education, female employment and family size, are not significant, but the result suggests a housewife will be more likely to be a deal prone when her age increases. An explanation offered by Webster (1965) is that younger women will be less expert in searching deals, thus, the search cost of a deal will rise. Likewise, McCanne (1974) analyses income, employment status of housewife, household size and age of housewife but find that they are non-significant.

Cotton and Babb (1978) analyse a wide range of demographic characteristics simply. In their research, geographical features, occupation and income level vary across different variables and do not show a significant relationship with deal proneness. Amounts of family members and educational level show negative relationship with deal redemption rate, but housewife employment status positively relates. More strikingly, non-white people are more likely to take the advantage of deals (Cotton and Babb, 1978).

Another study shows a family with children or employed wife will not take advantages of coupons due to these activities will cost a great amount of time (Blattberg et al. 1978). Additionally, Blattberg et al. (1978) analyses five different usually bought goods and finds that a larger percentage of households with cars and houses are deal prone than those without cars and houses. This empirical data shows household resource variables are effective predictors of consumer behaviour. Similarly, higher income households are more likely to deal prone. However, they found that the youngest children's age and the occupation of hostesses are less important to decide their behaviour relating to coupons. After that, Teel et al. (1980) reach a conclusion that members from larger family, with larger income and younger age, are probably use coupons. It implies that the finding of Teel et al. supports Blattberg et al. (1978) at the issue of income. Also, Teel et al. (1980) attempts to research on the two other variables, which are education and housewife employment, but they fails because the result is non-significant.

In order to support the hypothesis that coupon users are more price elasticity, Narasimhan (1984) tests the demographic features of coupon users. After analysing his data across various categories of daily use products, Narasimhan (1984) draws an opposite conclusion towards Cotton and Babb (1978) that education is significantly positive related to deal proneness. The opinion on income effect is similar with previous studies except Blattberg el at. (1978). Moreover, Narasimhan (1984) focuses on the influence of children in household and he indicates that children will increase the possibility to purchase with coupons. Meanwhile, Narasimhan's study (1984) reveals an indicator of coupon proneness, that is, consumers will face a trade-off when collecting the coupons. If the coupon-using cost is lower than the obtained saving, people tend to redeem the coupon and earn that benefit. Otherwise, consumers will prefer finding other saving methods.

Bawa and Shoemaker (1987a) collect seven product classes based on the unit of household and they imply that demographic characteristics are decent predictors of consumer redemption behaviour. Consistently with the hypothesis, less educated, lower income, suburb, smaller family consumers are less expected to be coupon proneness. The other three tested variables, which are housewives' personal features (age, employment and childbearing history), are non-significant in the test.

To summarize, the demographics test results vary from period to period and from researchers to researches. At a glance, the effect of income seems to be consistent in prior studies, Blattberg el al. (1978), Teel el al. (1980), Bawa and Shoemaker (1987a) suggest the positive relationship with deal/coupon proneness. The other scholars do not draw a totally opposite conclusion against these three articles. Moreover, geographical characteristics and presence of children are the field that less likely to be explored. Due to the different of dataset, researchers will continue to present different but inspiring conclusions in this topic.

Additionally, Mittal's (1994) attitude towards demographics is comprehensive. On one hand, Mittal (1994) insists that demographic indicators are poor when estimating the behaviour. He prefers to research on the reason coupons are used rather than the demographic characteristic of coupon users. On the other hand, he presents a causal model in his article (Mittal, 1994) that demographics is the mediation of consumer behaviour. It mediates subjective and domain individual-difference variables, even costs/benefits perception. Then, it contributes on coupon attitudes. However, he insists that demographics could not be the starting point of this process. Overall, it implies that future research should notice the influence of demographics, but cannot only concentrate on the demographic characteristics.

2.1.1.2 Subjective IDVs

In this field, the achievement of consumer redemption behaviour is less productive than demographics (Mittal, 1994). Subjective IDVs can be seen as psychological characteristics. Montgomery (1971) conducts a research on demographic, psychological features and purchasing behaviour (domain IDVs) and reveals that all variables of subjective IDVs are not significant at the whole study. Remarkably, gregariousness, venturesomeness and media involvement positively relates in the first phase of the research, but in the second phase, they become insignificant statistically. Also, McCanne (1974) involves testing creativity in his research. However, non-significant result cannot provide any convincible explanations for the relationship between psychological characteristics and deal-prone consumers.

2.1.1.3 Domain IDVs

Domain individual-difference variables are traits that derive from past purchasing experience. As the second highest attention topic in consumer coupon/deal redemption behaviour, past purchase behaviour is a more stable indicator of consumer behaviour according to the prior conclusions.

Webster (1965) uses two independent variables to access the brand loyalty, one is percentage of the loyalty brand products to total units and the other is numbers of different brands bought. The result of regressions shows that tendency to deal prone negatively correlates with the first variable and positively relates to the second one. This significant statistics data reveals that consumers are less likely to take the deals and change the purchased brand because of brand loyalty. It also implies that brand switchers, who have a high number of purchased various brands, are more likely to be a deal proneness. In addition, Webster (1965) examines the total purchased units as well. Contrast with the rational thinking, larger amounts purchasers are probably not deal proneness, proven by the statistics. Webster (1965) explains that large usage rate may result in less controllability of purchasing timing. In other words, bulk purchasers cannot wait until the deal appears because they need to keep the inventory level at a large amount. In addition, Webster (1965) also examines the other factors, such as the number of shopping trips, but these factors are not significant in his model.

Montgomery's study (1971) only has one significant variable, brand loyalty of Crest before endorsed by American Dental Association. It claims that loyal customers will negatively be a deal proneness. This idea supports Webster's (1965). Moreover, McCanne (1974) demonstrates that usage rate should be positively related rather than negatively related and loyalty factors will keep consumers away from deals. Nevertheless, data of McCanne is non-significant at the 5% significance level and leads to a less acceptable result.

A decade later, Shoemaker and Tibrewala (1985) indicate that most coupon-redeemed consumers have past purchased experience with the chosen brands. An implication can be suggested that more prior purchasing, more loyalty, thus more redemptions. Likewise, Bawa and Shoemaker (1987a) analyse not only demographic features but also domain IDVs. They claim that both brand loyalty index and store loyalty index are higher in the segment of non-coupon-prone customers than the segment of coupon-prone consumers. This result is accord with the previous research and strengthens them.

Another exception of consistent result is that Teel et al. (1980) support the negative impact of brand loyalty as well, but the data is surprisingly non-significant at store loyalty variable. Additionally, if a consumer is more sensitive to price, he or she will choose redeem a coupon (Teel et al. 1980).

Overall, the conclusions are consistently significant till now except the period of 1974~1980. It should be noticed that the research in those periods is not opposite but just insignificant statistically. Hence, a trustable conclusion can be drawn that brand loyalty can prevent consumers from redeeming coupons.

2.1.1.4 Cost/benefit perception

During recent three decades, the researchers transfer their focus on utility perception theory (Mittal, 1994). Shimp and Kavas (1984) argue that more money-saving and less time consuming when seeking a coupon lead consumers to coupon prone. Narasimhan (1984) concludes that the redeeming decision is based on the individual tradeoff between search costs and savings obtained. He directly relates the cost of coupon to the opportunity time of finding deals or coupons. In addition, Lichtenstein et al. (1997) shows that it is cost effective to only focusing on a certain amount of deals. If deal prone consumers try to target more, the cost will increase but the rise of savings cannot be guaranteed.

Moreover, due to the low price, consumers probably regard the underlying products as cheap, unqualified or inferior ones, thus the purchasing and redeeming behaviour are interrupted (Lichtenstein et al. 1993).

2.1.2 From the perspective of coupons/deals

Spotlights of redemption are almost on analysing the consumers performance, few scholars concentrate on the coupons/deals themselves before 1980s. Recently, academic research begins to focus on the characteristics adhere the coupons.

Firstly, face value of the coupons attracts the scholars. Traditional wisdom nearly reaches the same conclusion that higher face value is associated with higher redemption rate of coupons (Ward and Davis, 1978; Shoemaker and Tibrewala, 1985; Bawa and Shoemaker, 1987b; Bawa et al. 1997). However, the effect on redemption rate maintains in a same level once the face value reaches certain interval (Bawa and Shoemaker, 1987b).

Secondly, underlying product becomes a factor that influences the redemption behaviour. The research of Bawa et al. (1997) indicates that if the underlying product of the coupon is a product with high usage rate, it will be more attractive to consumers. Consequently, a large amount of users will redeem the coupon. Besides, in Gonul and Srinivasan's paper (1993), statistics significantly suggests that perception of coupon value refer to the brand of underlying products.

Lastly, redeem timing research promote the strategies depicted for managers. Under the framework of regret theory, Inman and McAlister (1994) hypothesize that the redemption rate will rise just before the expiration date because customers will regret if they fail to benefit from the coupons. They theoretically and empirically prove the hypothesis, which is in contrast to prior research that rate of redemption may be largest in the initial period (Bowman, 1980). This finding provides implications that managers can present various expire date to construct suitable couponing strategies. Furthermore, Gonul and Srinivasan (1996) get rid of the limitation of specific promotion, they consider long term vision and demonstrate that there is a tradeoff between redeem currently and wait till next period. Inventory, stockout cost and expectation of future coupons influence the redemption behaviour.

2.2 On the context of ecommerce

Traditional wisdom about the coupon redemption mostly analysed the offline market, in which coupons are based on paper version and distribute by mail, leaflets, TV advertisement, etc. Online market shares some identical features but speciality of ecommerce should be noticed when analysing the online group purchasing behaviour.

In terms of commerce, low search cost, trust, lock-in and novelty are four key factors that will affect the consumer purchasing behaviour.

2.2.1 Low search cost

Salop and Stiglitz (1977) introduce the concept of market segmentation by information. That is, informed consumers pay less by shopping at low-priced store or coupons while uninformed customers suffer a higher price by randomly selecting a store or not using coupons. More importantly, high-priced stores will lose their market shares gradually due to word-of-mouth transition of information, which will convert uninformed consumers into informed ones. Therefore, price discrimination in such case is temporal. Varian (1980) enhances this conclusion that randomized pricing strategy, which raises search costs, should be taken to remove the ability of customer learning.

Schmitz and Latzer (2002) analyse previous research and find that it is easier to obtain information in ecommerce than brick and mortar market. Bakos (2001) claims that ecommerce market with low search costs will facilitate intensive price competition. However, the price competition may be reduced by raising tacit collusion between online merchants (Varian, 1999, cited in Schmitz and Latzer, 2002; Kauffman and Wood, 2001). Additionally, low search cost effectively increases the ability of inter merchants scrutiny in ecommerce market.

One technology to reduce the online search cost is price comparison websites. However, the lower search cost mechanism does not always work. Baye and Morgan (2000) find that the prices shown on the price comparison websites are lower than merchants' own websites. They hypothesize that the price comparison website is a monopoly gatekeeper of information (Baye and Morgan, 2000). The firms have to pay a certain amount of fee to display their prices on the price comparison websites, in order to earn more click, the displayed prices will be reduced so that they can attract more consumers. Despite that, the special fees generated by ecommerce, such as shipping fees, may contribute to the inconsistence of price.

2.2.2 Trust

Urban et al. (1998, cited in Brynjolfsson and Smith, 2000) believes that credibility essentially affects internet marketing. Similarly, Brynjolfsson and Smith (2000) examine whether the e-market is frictionless and draw a conclusion that trust is still important because the internet market isolates participants spatially and temporally. Furthermore, Grabner-Kraeuter (2002) argues that the decisive variable for success of ecommerce is trust in the near future.

Empirically, consumer behaviour towards price comparison websites can explain the importance of trust. Smith and Brynjolfsson (2001) study 20,268 book buyers and find that they prefer larger brand sellers rather than lowest price, which price advantage is about $1.72. According to this statistics, brand can be seen as a proxy for credibility and quality for merchants. In that case, ecommerce has to encounter a more severe problem- trust rather than price elasticity.

2.2.3 Lock-in

A great amount of online merchants provide different kinds of programs, such as Amazon Prime, to lock the consumers in. Typically, online firms offer free credit when you spend a specific amount, which enhance the difficulty of switching. More specifically, the recommendation reward, free credit for recommending a deal, is set up by the online group purchasing websites. Lock-in generates brand loyalty and consumer trust, which are invisible assets of a firm (Amit and Zott, 2001). Consumer learning prohibits consumers from switching merchants (Smith, Bailey and Brynjolfsson, 1999). The data of Amit and Zott (2001) supports this opinion. Further, Amit and Zott (2001) indicate that consumers will be loyalty to the online merchants because of customization.

Another form of lock-in is based on interaction that it will be harder for brick and mortar to perform in the same way. Hagel and Armstrong (1997) argue that e-communities that created by online merchants effectively connect the consumers and sellers. This can be seen as a positive network externality, which is defined as 'the utility that a user derives from consumption of the good increases with the number of other agents consuming the good' (Katz and Shapiro, 1985, pp.424). Online group purchasing websites can simply reach this goal by designing a bulletin board system (BBS).

2.2.4 Novelty

Amit and Zott (2001) indicate that ecommerce innovate the business model and structure the business transactions against the traditional market. E-business market attracts new participants and depicts new relationship between products, services, information, seller and consumers (Amit and Zott, 2001).

Low search cost reduces the consuming time to search deals. More precisely, shopping guidance websites directly lead customers to suitable online group purchasing deals. Additionally, lock-in effect increase the brand and store loyalty. According to the review of consumer redemption behaviour, customers with low searching cost and more loyal tend to be deal proneness. By contrast, problems of trust and new system may prevent those less risky consumers from buying online. However, these results are based on logical reasoning followed the prior research, so the relationships are yet to be confirmed by data or evidences.

3. Hypotheses

A great number of previous literature show that the correlation between demographics and deal proneness remains controversial. So the demographics will be re-examined in this paper. According to results of literature, the assumptions are made as below:

H1: Gender affects the tendency to purchase online group deals. More precisely, female tends to capture a deal rather than male.

Traditional wisdom focuses on the household as a unit of observation, thus, the gender difference do not explain but the analysis of housewives' employment, education and age is the highlight of past studies (Webster, 1965; McCanne, 1974; Cotton and Babb, 1978; Narasimhan, 1984).

H2: Age negatively impacts on the deal prone. Older people are less skilled in computer and do not accepted the transaction style of ecommerce, which is a new shopping mode.

H3: A Bachelor degree holder probably purchase online group deals because of owning adequate computer skills.

Online group purchasing requires the basic knowledge of searching online and making payment via online banking. By contrast, the traditional coupons are distributed by easy accessible method. Therefore, age and education may be good indicators that the knowledge consumers own.

H4: The less monthly expenditure is, the more probable to purchase online discount deals. Those spending more monthly will do not restrict their purchasing behaviour by chosen deals.

This assumption seems adverse towards the previous findings (Blattberg et al., 1978; Teel et al. 1980; Bawa and Shoemaker, 1987a). They suggest higher income implies more controllable resources for taking the advantage of deals (Blattberg et al., 1978). However, according to Lichtenstein et al. (1993), consumers perceive price promotion products as inferior goods. In China, the 'face' (Hofstede and Bond, 1988, pp. 8) is very important especially in upper class. Consequently, the higher income class, which is probably the higher expenditure class, may regard online discount deals as inferior products, thus prefers to pay at regular price for the perceived superior products. Moreover, the variable, expenditure, reflect the purchasing value more exactly than the variable income. Thus, this paper chooses expenditure as a testing variable.

H5: Purchasing behaviour of other ecommerce forms, such as taobao (Chinese ebay), positively increase the purchasing times of online group deals.

Although online shopping did not develop in last few years, some consumers still biased against electronic business in China. In other words, if a consumer has not yet bought goods online before, he or she probably refuses to purchase such discount deals because of cheating in ecommerce. On the other hand, if a consumer is accustomed to shopping online, the likelihood that he or she purchases an online group deal augments. This behaviour can be defined as channel loyalty. Hence, we assume that the number of purchasing other forms of ecommerce has positive relationship with the amount of buying electronic group deals.

4. Research method

In order to explore the relationship between purchasing rate of online group deals and gender, age, education, monthly expenditure, frequency of shopping online, the author gathered information and data by sending questionnaires to candidates with various backgrounds.

The questionnaire contains six questions and collects the demographic information and the domain individual-difference variables. The variables, gender, education, frequency of monthly online group purchasing and monthly total online purchasing trips, were investigated directly. However, the age and monthly expenditure were shown as interval. Due to the convenience of dealing data, this essay used the midpoint of the interval as the data to run regressions.

The questionnaire was distributed by online and offline. On one hand, online survey was based on a famous Chinese online survey website Sojump.com. In order to increase the incentive to finish the questionnaire in high quality, the website offered lottery to earn a gift if the candidates repost the link of the questionnaire via social network services. Some experienced online shoppers were invited to enter the survey by spreading the questionnaire in Sina Weibo. Also, sample was randomly selected and invited to enter the survey by other instant message tools, such as QQ. On the other hand, offline survey was conducted to focus on the group who is not always online, for example, the older people and less educated people.

The survey collected 101 samples, including 61 women and 40 men. This study requires six variables. Firstly, gender. Gender is denoted as 'GENDER' in the STATA. It is a dummy variable. Gender=1 if the candidate is a male and gender=0 if the sample represents female.

Secondly, age data, denoted as 'AGE', was collected by the form of interval. These data will be analysed by the midpoints of the interval in ordinary least square regression. Thus, 21~25 age group will be simplified for 23 in the regression. This treatment of data may rises inaccuracy when data varies within the same interval. Due to the inaccessibility of precise data, it is an acceptable way to collect necessary data.

Thirdly, because the bachelor degree holders generally have adequate technical knowledge to purchase online, the education variable, denoted as 'BAC' in STATA, is also classified as a dummy variable. 1 represents holding a Bachelor degree, while 0 means the sample's educational level is below undergraduate.

Fourthly, monthly expenditure is treated as age. Midpoints of intervals are used in the regression. This variable is denoted as 'EXP'.

Lastly, numbers of purchasing online group deals within a month, denoted as 'GPT', and amounts of ecommerce transactions within a month, denoted as 'EPT', are directly collected the exact number of each candidate. Because of the inclusion of GPT, a new variable is generated by subtracting GPT from EPT, which is denoted as 'NEPT'. This new variable describes the purchasing quantities of ecommerce, net of group-buying.

To measure the relationships between these variables, regression analysis will be presented. This essay will explore from the simple linear regression model, multiple regression to the non-linear regression model. The tests process and results will be shown in the next part.

5. Data analysis

5.1 Correlation analysis

Firstly, in order to examine the relationships between the variables, a correlation analysis will be conducted. As shown in the Table 1, only two variables, gender and NEPT, significantly relate to the purchasing times of group-buying deals (GPT). More precisely, female are more likely to generate a large number of online group deals (p-value =0.0314<0.05). Also, the amount of purchasing online, net of online group-buying, is positively related with the numbers of purchasing groupon deals. The correlation between GPT and NEPT is highly significant (p-value =0.0000<0.01) at 1% significant level. Therefore, two hypotheses, H1 and H5, are supported by the data causally.

Table 1: Results of correlation analysis

GENDER

AGE

BAC

EXP

NEPT

GPT

GENDER

1.0000

AGE

0.1581

(0.1143)

1.0000

BAC

0.0471

(0.6401)

-0.5640

(0.0000)*

1.0000

EXP

0.0122

(0.9037)

0.3682

(0.0002)*

-0.2436

(0.0141)*

1.0000

NEPT

0.0106

(0.9166)

-0.1095

(0.2757)

-0.1166

(0.2457)

0.0655

(0.5152)

1.0000

GPT

-0.2143

(0.0314)*

-0.0262

(0.7948)

-0.0567

(0.5731)

-0.0538

(0.5932)

0.4774

(0.0000)*

1.0000

* p-value <0.05, the data is significant at 5%.

In terms of the other three hypotheses, from the Table 1, the results of correlation analysis do not provide significant correlation between those variables. Specifically, the p-values of correlation between GPT and AGE, GPT and BAC, GPT and EXP are 0.7948, 0.5731, 0.5932, respectively. In that case, these three hypotheses should be rejected at the significant level of 5%.

Additionally, the correlation between age and whether holds a Bachelor degree is slightly high (-0.5640, significant at 1%). It will possibly result in imperfect multicollinearity when choosing these two variables in the same regression model. In the hypotheses part, these two variables are chosen to indicate the same effect, meeting the technique requirement of purchasing online. Thus, there will be possibly some internal relationships between these two variables.

5.2 Regression analysis

Secondly, regressions are run to explore the linear or non-linear relationship between variables. Overall, as shown from the Table 2, the R2 of Equation 1 is relatively higher than dropping some variables such as AGE and BAC. However, it does not imply that the additional variables exactly increase the fit of the regression model. After analysing the adjusted R2 (0.2594, lowest of all multiple regressions excluding non-linear regressions Equation 6) the fit of the regression even decrease the adding variables, AGE, BAC and EXP, are non-significant at 5% in the regression Equation 1 and Equation 2. Furthermore, the Table 2 also reveals the most suitable model of online group deals purchasing behaviour analysis. Both equation 3 and 5 provide significant coefficients (all p-value<0.01, significant at 1%) of all independent variables. However, the R2 and adjusted R2 of Equation 3 are relatively lower than the ones of Equation 5. Thus, Equation 5 is a more precise model to demonstrate the relationship between group deal proneness and its decisive factors.

Table 2: Results of regressions (N=101)

Equation 1

Equation 2

Equation 3

Equation 4

Equation 5

Equation 6

GENDER

-1.010663

(0.006)***

-0.8967914

(0.009)***

-0.9007049

(0.008)***

-1.061752

(0.003)***

-0.9525089

(0.006)***

-0.7377234

(0.053)*^

AGE

0.0381689

(0.350)

BAC

0.4471700

(0.715)

EXP

-.0000672

(0.099)*

-0.0000449

(0.126)

NEPT

0.4891361

(0.000)***

0.4620995

(0.000)***

0.4569375

(0.000)***

0.845538

(0.000)***

0.9891617

(0.000)***

0.4732911

(0.004)***

NEPT2

-0.0489191

(0.055)*^

-0.0612083

(0.005)***

GENDERÃ-NEPT

-0.0719672

(0.736)

CONSTANT

-.5227027

(0.801)

.8225716

(0.004)***

0.6861721

(0.005)***

0.3410535

(0.159)

0.6497116

(0.025)**

R2

0.2965

0.2829

0.2761

0.3073

0.5191

0.2771

Adjusted R2

0.2594

0.2607

0.2613

0.2859

0.5043

0.2547

RMSE

1.7365

1.735

1.7343

1.7052

1.7059

1.742

* p-value <0.10, the coefficient is significant at 10%.

** p-value <0.05, the coefficient is significant at 5%.

*** p-value <0.01, the coefficient is significant at 1%, highly significant.

^ p-value is just over 0.05, the coefficient is approximately significant at 5%.

More exactly, Equation 1 indicates the relationship between tendency to purchase online group deals and all assumed decisive variables. GENDER (p-value=0.006<0.01) and NEPT (p-value=0.000<0.01) are significant at 1% but AGE (p-value=0.350>0.10) and BAC (p-value=0.715>0.10) are non-significant. Hence, the H0 of coefficients (βage=0 and βbac=0) cannot be rejected even at 10% significance level and these two coefficients probably equal to zero. Expenditure is significant at 10%, thus, it is reasonable to keep this variable into the next regression and drop AGE and BAC. In that case, if holding other variables constant, there is no evidence that older people will increase the purchasing amounts of online group deals (H2 rejected). Likewise, Bachelor degree holders cannot be proven to be more likely to take advantage of group deals (H3 rejected). By contrast, this result claims that men are approximate one time less purchasing than women (H1 accepted). Also, the result shows that NEPT is positively related with GPT (H5 accepted). It shows a negative correlation between monthly expenditure and GPT, but the 10% significance level suggests this relationship should be stated carefully (H4 accepted at 10% significance level). We should notice that the constant is non-significant as well, so in the following regression, dropping the constant should be taken into consideration. Lastly, the relatively high root mean squared error of Equation 1 suggests that this regression cannot predict the data as well as the other regressions which have a smaller RMSE.

In Equation 2, following the result of Equation 1, the regression drops some variables proven non-significant in the prior regression. Consequently, the adjusted R2 increases by 0.0013 and reaches 0.2607, which suggests the fit of regression is better than Equation 1. In terms of individual variables, GENDER and NEPT are consistent with the previous result, significant at 1%, which indicates the H1 and H5 are supported. However, the coefficient of monthly expenditure even cannot be rejected at 10% significance level. Moreover, the coefficient of EXP is very close to zero, amounting at -0.0000449. As seen from the Table 1, EXP is significantly correlated with AGE and BAC. The removal of AGE and BAC may affect the performance of EXP. Hence, according to this analysis, the Hypothesis 4 is supposed to be rejected at the 10% significance level. The effect of variables increased by a unit is similar with the Equation 1. Last but not least, constant of this regression is different from the Equation 1 and significant at 1%, thus, the constant of Equation 1 is non-significant probably because BAC is a binary variable that has substantial influence upon the constant.

One reason for the inconsistency of monthly expenditure may be due to the interval data. EXP is collected by interval and used the midpoints of the interval as the specific data when running ordinary least squared (OLS) regression. The uncertainty within the interval cannot be detected and analysed. Therefore, the result may be impacted by this type of data processing.

Equation 3 provides a better predictable result than Equation 1 and Equation 2 due to the drop of non-significant variable EXP. All coefficients (p-value of GENDER=0.008<0.01; p-value of NEPT=0.000<0.01), including constant (p-value=0.005<0.01), are significant at 1%. The adjusted R2 (0.2613) is higher than the other two regressions, while the root mean squared error (1.7343) is lower than Equation 1 and Equation 2. Although the adjusted R2 is not large, the significance of coefficients implies that the result is reliable. This regression suggests the same result with Equation 2, supporting the H1 and H5.

The small R2 of Equation 1, Equation 2 and Equation 3 indicate the linear model may be not suitable for this analysis. Two reasons can explain this phenomenon. Firstly, there are some omitted variables existing. The factors that influence the consumer purchasing behaviour may be existed except the GENDER and NEPT, such as the frequency of being advertised about the online group deals. Secondly, linear relationship may be not effective to describe the data. Hence, non-linear regressions will be conducted to explore the relationship between different variables.

A quadratic regression model is applied in Equation 4. First of all, generate a new variable in STATA that NEPT2, which is the square of NEPT. According to the Table 2, the result of Equation 4 is more adequate explaining the variables than Equation 3. The adjusted R2 enhances and the RMSE decreases. The quadratic term is approximately significant at the significance level of 5%, which shows that it is reasonable to generate a quadratic term. However, the constant term is non-significant even at 10%. Thus, the constant term will be dropped in the next equation.

Equation 5 is a revised version of Equation 4. All terms (GENDER: p-value=0.006<0.01; NEPT: p-value=0.000<0.01; NEPT2: p-value=0.005<0.01) are significant at 1% according to the test results of t-statistics. Surprisingly, the R2 (0.5191) and adjusted R2 (0.5043) are substantially increase comparing with the previous regressions, which means that the data is better described in this quadratic regression model. Due to the high R2 and all significant coefficients, this equation will be chosen to develop the model of consumer behaviour.

Simply holding NEPT constant, thus NEPT2 constant, GENDER shows a negative relationship with GPT. Specifically, GENDER is a binary variable (male=1, female=0) and negative correlation means female are more likely to be a group deal proneness, which strongly supports H1.

In terms of H5, the relationship is more complicated than prior research. The graph of Equation 5, regarding the GENDER is a constant, is a downward parabola. When NEPT equals 8.08, it reaches the peak of the parabola. In that case, when NEPT is less than 8.08, the relationship between NEPT and GPT is positive. GPT will increase followed by the increase of NEPT. However, after the peak, their relationship becomes negative. That means, once consumers purchase more than 8 times from Internet, net of groupon deals, they will probably tend to be less incentive to the online group deals. Thus, H5 should be conditionally accepted because it describes the trend of first part of parabola.

In order to test whether the prediction of GPT is the same for male and female, an interaction term should be constructed between the binary variable and explanatory variable. Thus, interaction term, GENX, equals GENDERÃ-NEPT, is generated in STATA. Another regression will be run based on the Equation 3 by adding a new term, GENX.

In Equation 6, let gender=1, the equation can be rewritten: Equation 7: GPT=0.402Ã-NEPT-0.088. Let gender=0, the equation can be transformed: Equation 8: GPT=0.473Ã-NEPT+0.650. When holding NEPT constant, Equation 8 must be larger than Equation 7. That implies that female consumers have a larger purchasing rate (H1 accepted). In terms of H5, both Equation 7 and Equation 8 show a positive relationship between those two variables, thus, H5 cannot be rejected.

However, statistically non-significant coefficient (p-value=0.736>0.1) implies that there is no difference between the purchasing behaviour across gender. The R2 is lower than Equation 3 but RMSE is higher, which indicates the fit of the regression is worse than the prior equations.

Moreover, the significance level of GENDER increases to 10% when adding the interaction term. F-statistics test of the joint hypothesis that GENDER and GENX equal zero shows that the null hypothesis can be rejected at the 1% significance level (p-value=0.0075<0.01). Thus, drop the interaction term and keep the variable GENDER is a more appropriate data processing method.

Non-linear regression model includes logarithm models, which are linear-log model, log-linear model, log-log model. The data of GPT and NEPT contains large numbers of the figure, zero, using logarithm model will reduce the sample size and result in unreliable results. Consequently, logarithm model will be abandoned in this circumstance.

To sum up, although some data are not significant statistically, two hypotheses (H1 and H5) are tested and proven (H5 conditionally accepted). EXP is only significant in Equation 1 at 10% significance level, so H4 cannot be accepted and some further evidences must be found to test this hypothesis. Due to the non-significant data, H2 and H3 are rejected statistically. Moreover, the best model to describe the consumer behaviour is Equation 5, which implies that the correlation between GPT and NEPT is non-linear.

6. Discussion and model of consumer behaviour

6.1 Female are more active to capture a deal

According to the six regressions from the last part, H1 can be accepted with a high significance level. General observation shows that female may be more likely to seek deals if they have enough time and money, but men possibly only go shopping when they need some products. Gender difference influences the behaviour difference inherently.

Previous consumer redemption behaviour studies (Webster, 1965; Cotton and Babb, 1978; Naramsihan, 1984) mainly choose household as a unit of data collection. The gender difference is not in their concern. However, in the unit of household, prior studies (Webster, 1965; Cotton and Babb, 1978; Naramsihan, 1984) concentrate on the housewives' behaviour, by testing the housewives' age and employment, rather than males'. It partially supports that the gender difference actually affect the consumer behaviour.

Moreover, this result can be supported by the previous research on gender issue. The studies analysed the issue of gender difference indicate that female are more likely to participate a purchasing activity than male in brick and mortar. Bergadaa et al. (1995) survey over 700 participants in four aspects and show that female are higher tendency to shopping than male, especially in leisure and social aspects. Dittmar and Drury (2000) point out that purchasing is a psychological and emotional desire for women. Additionally, Campbell (2000) demonstrates that male only purchase when they need and try to minimize consuming time and effort on shopping. Thus, it is reasonable to draw a conclusion that gender difference affects the consumer behaviour and female are active deal proneness.

However, the previous studies on online market of purchasing probability are not consistent with the brick and mortar market (Dittmar et al. 2004). Scholars mainly reach the conclusion that the volume and frequency of purchasing by male consumers is larger than those by female (Li et al. 1999). Li et al. (1999) suggest that gender difference of online purchasing is due to the inequal knowledge of technology. Traditionally, men master computer technology better than women. In that case, women purchase less online because they hardly know how to purchase online. Nevertheless, the study (Li et al. 1999) was conducted early from nowadays. In the recent decade, the market of ecommerce becomes more mature and expands larger, while the proliferation of computers equip most people the basic ability to use computer. After that, accessibility is no longer a major problem for large proportion of customers. Female are capable of buying online as male, especially young women. Hence, the result of this analysis is inconsistent with the previous studies but reasonable due to the improving society.

Swaminathan et al. (1999) argue that men online consumers prefer online purchasing because they are convenience-oriented and avoid social activities. It seems to reject the Hypothesis 1 by suggesting the male consumers are more likely to choose the channel of ecommerce. Considering the advanced society as last paragraph, this argument should be revised. By solving the problem of lack of technology, women can purchase online as convenient as they purchase in conventional channels. Thus, the traditional assumption that women tend to shopping is applied again. This explanation suggests that the H1 can be accepted under the current society. Consequently, combining the idea of Swaminathan et al. (1999) with the current situation, the gender difference is not so obvious as traditional channels because the ecommerce channel improve the attractiveness to male consumers by offering a convenient and less social shopping style. Statistics supports this assumption by showing the gender difference of monthly purchasing online deals is just approximate 1 (see Table 2).

Lastly, in China, traditional ideology that men work outside and women pay more attention to the household daily living may help explain this relationship. Under that concept, women are more sensitive to saving money and planning daily life, such as purchasing daily goods. They will be more likely to take the online group deals when they are researching the information about purchasing daily goods. Gender's effect is consistent in all regressions in this analysis and can be explained reasonably, thus, H1 is accepted.

6.2 Age and education are non-significantly correlated with online group deal prone

Recall from literature review, McCanne (1974), Bawa and Shoemaker (1987a) both reveal that the relationship between age and deal prone is non-significant, which is the same conclusion as this paper. H2 hypothesizes that aged people are incapable of using computers and searching the internet. As mentioned in 6.1, accessibility is probably not a major issue for shopping online. That is, computers and basic online purchasing ability are spread widely, except those are over 60 years old. More and more people in the age group of 40~50 are capable of basic knowledge of using high-tech products. Thus, this paper can reach a conclusion that the relationship between age and deal prone is non-significant.

However, this relationship is inconsistent in traditional wisdom. Webster (1965) argues that older consumers are equipped with powerful ability to search deals in the market. Considering the difference between online and offline market, it is reasonable to reach a conclusion that is slightly different from Webster. In reality, older people may show hostility against web shopping due to the spatially separated transaction environment, but young men are more tolerate especially those who tend to stay at home. Thus, due to the market difference, Webster's finding (1965) should be revised to fit the current situation. Nevertheless, the statistics (p-value of AGE>0.1 in Equation 1) do not show any evidence about this assumption. Teel et al. (1980) suggest age negatively impacts the purchasing frequency, which is adverse to Webster. The diversified literature results cannot provide a universal explanation about this relationship, thus, it is safe to find out that the age may be not the determinant factor of deal proneness (H2 rejected).

Likewise, H3 suggests that higher educated consumers know the basic knowledge of computers. The p-value of BAC in Equation 1 is 0.715 (see Table 2), which indicates that the data is non-significant at even 10% significance level. This finding is consistent with the results of Webster (1965) and Teel et al. (1980). An explanation of this non-significant data is that Bachelor degree is not a good predictor to describe the computer skills of a participant. In China, many youngsters are addicted to play computer games so that they fail to enter a university. In that case, this type of consumers is capable with the ability to purchase online without a Bachelor degree. To describe the accessibility of ecommerce, predictors, such as computer skills and online banking holders, are more suitable to predict.

However, recent studies show that the positive relationship between education and deal proneness is significant (Narasimhan, 1984; Bawa and Shoemaker, 1987a). On one hand, the positive coefficient (0.4471700, see Table 2) shows that the relationship between these two variables in this dataset is positive, which is accordance with the conclusion of Narasimhan (1984), Bawa and Shoemaker (1987a). On the other hand, the non-significant result implies that the positive correlation is not trustable, which is inconsistent with the above two results. The inconsistency with several literatures (Narasimhan, 1984; Bawa and Shoemaker, 1987a) but consistency with other literatures (Webster, 1965; Teel et al., 1980) may be due to the random sampling. Hence, a more persuasive analysis needs to be conducted in the future.

Overall, statistical findings demonstrate that age and educational level of the consumer have no significant correlation with purchasing tendency (p-value of AGE>0.1 in Equation 1; p-value of BAC>0.1 in Equation 1). These results imply that H2 and H3 can be rejected. It is reasonable to draw such a conclusion because the prior studies can hardly reach a consistent conclusion about these two variables, age and education.

6.3 Inconsistency result of monthly expenditure

The conclusions of H4 are different in two regressions, unlike H1, H2 and H3. Equation 1 shows monthly expenditure is negatively correlated with GPT at 10% significance level, while Equation 2 indicates negative relationship without significance.

Past research (Webster, 1965; McCanne 1974; Cotton and Babb, 1978; Blattberg et al., 1978; Narasimhan, 1984; Teel et al., 1980; Bawa and Shoemaker, 1987a) mainly analyse the variable income rather than expenditure. Generally, income is a predictor of expenditure, larger income results in larger expenditure. Early studies (Webster, 1965; McCanne 1974) are consistent with the result of Equation 2, showing the variable coefficient non-significant. The interval data, which does not reflect the fluctuation inside the interval, may be a reason that the data reach a non-significant result.

However, the results of previous studies are various. Despite the confused results, scholars draw a conclusion that the correlation is positive (Blattberg et al., 1978; Teel et al., 1980; Bawa and Shoemaker, 1987a). Previous studies infer that higher income allows consumers more controllable household resources to purchase more deals (Blattberg et al., 1978). Nevertheless, negative relationship derived in Equation 1 cannot be supported by the previous research. Culture difference between eastern and western society should be introduced to explain the difference between this paper and traditional wisdom. When the assumption is made, this article mentions that Chinese people value their 'face' (Hofstede and Bond, 1988, pp. 8), which results strongly in consuming the famous brand products and luxuries, is more important than other indicators. In the framework of Confucian Dynamism value, protecting one's face is an essential value to live in an eastern society, such as China (Hofstede and Bond, 1988). In that case, spending money on luxury products can protect one's face and receive respect. Moreover, price promotion products are perceived as inferior goods (Lichtenstein et al., 1993). Consequently, the online group deals mostly are not in the category of high-valued and luxury goods. Thus, the high expending consumers will choose other valuable products to protect their faces rather than discount deals. Wealthy customers are less likely to choose a saving instrument, so the monthly expenditure variable is negatively related with the GPT.

Additionally, expenditure is slightly different from income. The variable expenditure also indicates the amount the consumer spent. If the two consumers purchase the same products but spend different money, the lower expending one may be use online group deals for saving money, while the other one purchases at the regular price. Thus, the expenditure variable directly reflects the purchasing time of online group deals and this relationship should be negative.

Above all, H4 cannot be accepted because of inconsistency in two regressions and insignificance at 5% significance level in both regressions. Nevertheless, if the significance level increase, it is reasonable to cautiously accept the assumption that the negative relationship between EXP and GPT due to the culture difference explanation.

6.4 Channel loyalty keeps the consumers

The strongest relationship among these variables is NEPT and GPT. The p-value of the NEPT is very close to 0.000 at any regressions, which implies that the coefficient of NEPT is significant. Equation 5 reveals the relationship between these two variables is downward parabola. Descriptions should be divided into two parts, before peak and after peak.

Before peak point, the correlation between NEPT and GPT is positive, thus, this result provides strong support for the channel loyalty. NPET is defined as the frequency of online shopping excluding the online group purchasing, and GPT is the frequency of online group purchasing. The significant relationship of these two variables indicates that online shoppers are more likely to choose a deal that displays on the Internet. This can be defined as channel loyalty. Recall from the literature review, brand loyalty and store loyalty keep the customers away from the promotion products (Webster, 1965; Montgomery, 1971; Teel et al., 1980; Bawa and Shoemaker, 1987a). Similarly, channel loyalty keep the consumers choosing the promotion sold online rather than other channels. That explains why customers with higher online purchasing experience will purchase more online group deals.

Inactive online shopping consumers may be loyal to their accustomed channel and hesitate that whether shopping online is a safe way due to the trust problems of online shopping (Brynjolfsson and Smith, 2000). Moreover, it is important to notice that the online shopping market in China is unregulated and a great amount of financial crimes are reported because of the lack of trust in ecommerce in China. For instance, several online shopping purchasing websites are complained for selling products distinct from displayed. Also, newspaper reveals that a new trend of unethical job is rising in China, which helps online retailers to harass consumers who report negative feedback. These facts and theoretical explanations suggest that infrequent buyers will prefer other type of promotion rather than online group deals, so that they can make sure the products or services are the ones they want.

Furthermore, online shopping is a non-traditional shopping channel and shares a special characteristic, lock-in. Lock-in programs can generate effects like brand loyalty (Amit and Zott, 2001). In other words, online shopping can easily construct the loyalty system of sticking consumers. That explains why a consumer with a higher NEPT will tend to be group deal proneness. More precisely, a consumer who is loyal to shop online may not change his behaviour to accept offline discount, he will prefer to use online discount, such as online group deals. This effect is similar with the effects of brand or store loyalty indicated by Webster (1965).

However, after the peak, the quadratic term suggests that the tendency to purchase online group deals will decrease. If a consumer purchases online over 8.08 times per month, he or she may have limited controllable resources remaining to take the advantage of other deals, such as groupon. Additionally, if a consumer with high NEPT but low GPT, it implies that he may be interested in purchasing other forms of ecommerce and loyal to that channel. Moreover, considering that 8.08 times per month is a relatively high frequency which is not usually reached by too many people, the relationship between NEPT and GPT can be roughly seen as a positive correlation in common situation.

Therefore, H5 is conditionally accepted and an amendment should be attached. That is, positive correlation between purchasing