Comment on Gruntkowski, L.M.; Martinez, L.F. Online Grocery Shopping in Germany: Assessing the Impact of COVID-19. J. Theor. Appl. Electron. Commer. Res. 2022, 17, 984–1002

Abstract

Gruntkowski and Martinez examined the impact of factors such as perceived risk and perceived usefulness on German consumers’ intention to purchase groceries online once the COVID-19 pandemic had subsided. They also compared consumer perceptions before and during the COVID-19 outbreak. This comment shows that Gruntkowski and Martinez’s research suffers from a number of problems, the most important of which is the use of an unrepresentative sample. They should therefore have refrained from generalizing their findings to the German population.

1. Introduction

In a recent article in this journal, Gruntkowski and Martinez [1] assess how COVID-19 has affected German consumers’ perceptions and intended use of online grocery shopping (OGS). Unfortunately, their research has multiple flaws. Before I demonstrate these, I first briefly summarize their approach and main findings.

2. Gruntkowski and Martinez’s Research

Gruntkowski and Martinez collected their data by means of an online survey conducted in October–November 2021 (i.e., during the third COVID-19 wave in Germany). Their questionnaire first asked respondents whether they had previously purchased groceries online. Those who had were subsequently asked about the timing of and reasons for their initial adoption, as well as their purchase frequency. All respondents also had to evaluate, on a five-point Likert scale, six “customer-oriented factors” [1] (p. 984) that could influence consumers’ intention to shop for groceries online (perceived risk, perceived ease of use, etc.). They were requested to do so twice: “The survey consisted of two main sections, corresponding to two different time periods—before and since the beginning of the COVID-19 outbreak” [1] (p. 990). A final construct was ‘Intention to shop groceries online’, which was obviously evaluated only once.

After data cleaning, Gruntkowski and Martinez retain 402 valid responses. With these data, they then first calculated Pearson correlations between all constructs, after which they explain purchase intention by means of simple and, ultimately, multiple linear regression. In addition, using paired-samples t-tests, they examine whether perceptions of OGS have changed significantly on account of COVID-19.

They find that the pandemic led to a strong reduction in perceived risk, whereas the ‘positive’ constructs (perceived usefulness, convenience, etc.) increased “at moderate levels” [1] (p. 984). Furthermore, all hypotheses concerning the impact on purchase intention are claimed to be supported.

3. Methodological Problems

Unfortunately, as mentioned, Gruntkowski and Martinez’s research suffers from multiple problems.

For one, in terms of internal validity, it can be questioned whether one can reliably measure consumers’ pre-pandemic perceptions of OGS in October–November 2021, that is, more than one year and a half after the facts. This would seem particularly problematic for the new adopters, who in the meantime have had firsthand experience with the pros and cons of online grocery services. These new adopters make up 52.3% of the sample [1] (Table 1, p. 990).

Second, surprisingly, Gruntkowski and Martinez do not wait for their multiple regression model to evaluate their Hypotheses 1a–6a about the impact of the individual constructs on purchase intention. They already confirm them based on Pearson correlations and simple linear regressions, and subsequently gloss over any differences between the univariate and multivariate results, see below.

Third, the research lacks an overarching theoretical framework. Constructs and scales are taken from a variety of models and sources, see Gruntkowski and Martinez’s Appendix A. As a result, there is overlap between several constructs and the regression analysis predictably suffers from multicollinearity.

In particular, the combination of perceived risk and perceived trust in one and the same model is surprising. The connection is already clear from the authors’ justification of the presence of the constructs. Concerning perceived risk, they write: “Perceived risk includes several factors such as personal data security, delivery issues, a lower quality of products than expected […]. With regard to data security, research emphasized that consumers have transaction risks when purchasing online as well as privacy concerns” [1] (p. 986). For perceived trust they write: “risks linked to payment, product, information and time impact consumers’ intention to purchase online negatively […]. Consumers cannot check the quality of a product physically or monitor the safety and security of sending personal and financial information while purchasing online” [1] (p. 988).

The problem becomes even more apparent when looking at the items of the respective scales. Item 2 of the scale for perceived risk is: “Security around payment and personal data on the internet is not good enough” and items 4–5 of the perceived trust scale are: “I feel safe using my credit card making grocery purchases online”, and “I feel safe to share my personal details if requested” [1] (p. 998).

There is also overlap between perceived usefulness and convenience. Item 1 of the scale for the former reads: “Shopping for groceries online increases my shopping productivity (e.g., I can use the time gained for sth. else)”. The similarity with items 1 and 3 of the convenience scale is evident: “Buying groceries online is time-saving” and “Delivery of the products to the door step saves time and physical exertion” [1] (pp. 998–999).

Given the above, it is not surprising that perceived risk and convenience are significant in the simple linear regressions but no longer in the full model, see [1] (Table 4, p. 993). To be fair, the authors do acknowledge that there is a multicollinearity issue: “it should be noted that the predictor perceived usefulness especially shows tendencies towards multicollinear structures in connection with the predictors ease of use and trust” [1] (p. 993). However, this remark relates solely to the three constructs that remain significant in the multivariate analysis. Nothing is said about the three other constructs—perceived risk, convenience, and situational factors—that are no longer significant. Note also that “[c]ontrary to expectations and the previous results” (that is, the simple linear regressions), according to the full model, ease of use would have a negative effect on purchase intention [1] (p. 993).

The authors nevertheless maintain that all hypotheses are supported. To quote from the Abstract: “The results indicate that perceived risk still has a negative influence on purchase intentions, thus remaining relevant in online grocery shopping. […] Moreover, perceived usefulness, perceived ease of use, perceived trust, convenience, as well as situational factors were found to have a positive relationship with purchase intention, both before the COVID-19 crisis and since then” [1] (p. 984). Note also that no such evidence is presented concerning the period “before the COVID-19 crisis”; Intention to shop groceries online is evaluated only once, namely in 2021.

The fourth comment is that Gruntkowski and Martinez never make a distinction between users and non-users, whereas the two groups, in fact, face a substantially different decision—respectively, to continue buying groceries online and to start doing so—and most probably have different perceptions. A recent study by Van Droogenbroeck and Van Hove validates an extended UTAUT2 (Unified Theory of Acceptance and Use of Technology) model for a sample of Belgian supermarket customers. They find that “users and potential adopters differ significantly on all constructs and that users have more positive views towards online grocery shopping” [2] (p. 12). The largest difference is observed for behavioral intention. As a result, Van Droogenbroeck and Van Hove also obtain different regression results for the (sub)samples they examine. For example, while perceived risk is significant in the full sample, this is not the case in the users and potential adopters samples [2] (Table 7, p. 17).

Note also that Gruntkowski and Martinez’s scale for Intention to shop groceries online would seem ill suited for non-users, containing as it does an item that reads: “I plan to continue buying groceries online once the COVID-19 situation has subsided” [1] (p. 999; my emphasis).

My final (and most important) remarks relate to Gruntkowski and Martinez’s sampling approach. The first observation is triggered by the recommendation in the literature [3,4] to target so-called primary grocery shoppers—the household members that are responsible for the majority of the grocery shopping—because otherwise respondents might be “questioned about an activity that they take no (or little) part in” [3] (p. 244). Gruntkowski and Martinez do not do this and rely on non-probability sampling: “The survey link was shared via the authors’ Facebook accounts, in relevant WhatsApp groups as well as via e-mail to reach as many potential respondents as possible” [1] (p. 989).

However, on closer scrutiny, Gruntkowski and Martinez’s sample may well comprise mainly primary shoppers. An indication in this direction is the dominance of women (62.2%) [1] (Table 1, p. 9), which is also typically the case in studies that do target primary shoppers. As Hansen [5]—in an early survey in the US—notes, his focus on primary shoppers “resulted in a majority of women participating…—a commonly detected tendency in studies of household grocery-shopping behavior” [5] (p. 108). Grocery shopping is indeed still a gendered task in many a country, with the possible exception of the Nordics [3] (pp. 256–257). Note also that the 62.2% in Gruntkowski and Martinez’s sample tallies quite well with the multi-country average proportion of females (60.5%) obtained by [3] (pp. 252–253) in their systematic literature review.

A possible explanation for the high proportion of women in Gruntkowski and Martinez’s sample, in spite of the fact that being a primary shopper was not an eligibility criterion, may be that (male) non-primary shoppers simply did not see any interest in participating. Interestingly, in a Dutch study on the relationship between online and in-store shopping for non-daily products, Farag et al. [6] observed that even though in the first stage of the data collection, households were selected randomly, women nevertheless form the majority (61%) in their sample. According to Farag et al., “A possible explanation […] is that shopping appeals more to women than to men. Hence, women would be more willing to fill out a questionnaire about shopping than men” [6] (p. 130).

This said, Gruntkowski and Martinez’s sample is unrepresentative in two other respects. For one, at 69.4%, the 20–35 age group is clearly overrepresented. Only 0.8% of the respondents are over 65 [1] (Table 1, p. 990). When the authors list the limitations of their research, they do mention that “a larger and more inclusive sample could potentially gain deeper knowledge of the subject. In [our] study, almost 70% of the participants are between 20–35 years old, which reduces its comparability to older age groups” [1] (p. 989). But mentioning the problem obviously does not make it go away. Crucially, if one knows that one has an unrepresentative sample, one should not draw general conclusions, see below.

Compounding the problem is that the sample also suffers from an overrepresentation of users of online grocery services. In their introduction, Gruntkowski and Martinez themselves write that while “the pandemic has brought a boom to online grocery shopping […]“, “[n]evertheless, in comparison to product categories such as fashion, consumer electronics and books, OGS is still a niche” [1] (pp. 984–985). Yet, no less than 71.9% of their respondents have purchased groceries online at least once [1] (Table 1, p. 990).

Although the figure is not directly comparable, note that according to Eurostat data for 2021, in Germany, only 10% of individuals had purchased food or beverages from online stores or from meal-kit providers in the previous three months [7]. This also happens to be the EU average. The highest number is for the Netherlands, at 28%. The best point of comparison would seem to be Appinio and Stryker’s ‘German Online Grocery Report 2022′ [8], which is based on a representative nationwide survey among 2500 consumers between 16 and 65 years of age. The January 2022 edition puts the proportion of German consumers buying at least some of their groceries online at 33%.

The implications of the described lack of representativeness are far-reaching. For example, the authors make much of their finding that perceived risk has dropped significantly compared to the pre-COVID setting. This may be true for the young users of online grocery services that dominate their sample—several of whom started using such services during the pandemic and were thus able to ascertain that their concerns were exaggerated or unfounded—but it is not a given that this is also the case for that part of the population that has so far decided against adoption. Also, the relationships between the constructs and purchase intention may well prove different in a representative sample.

In short, Gruntkowski and Martinez should have refrained from generalizing their findings to the German population. Yet their article is rife with general statements. This is a selection (the emphasis is always mine): “the consumers’ perceived risk is considered lower compared to the pre-COVID-19 scenario” (Abstract); “The insights gained from this study can help grocery retailers to respond appropriately to consumers’ expectations and reservations” (p. 985); “The results show clearly that German consumers are willing to buy groceries online, as the advantages of the online channel outweigh those of stationary shopping for them” (p. 995); “This shows that the ease of unwanted item returns are important to online shoppers when considering OGS” (p. 995); “The results show that [online grocery shopping has] become more socially accepted in Germany since the outbreak of COVID-19” (p. 996).

4. Conclusions

This comment has shown that Gruntkowski and Martinez’s research suffers from several problems: the internal validity of their measurement of respondents’ pre-COVID perceptions can be questioned, they accept hypotheses solely based on simple linear regressions (and ignore conflicting results of their multiple regression), the scales of specific constructs overlap and their multiple regression model suffers from multicollinearity, they fail to make any distinction between the users and non-users in their sample, and, most importantly, their sample is unrepresentative in terms of both age and use of OGS.

The upshot is that Gruntkowski and Martinez’s results are, at best, only valid for their sample and should be used with considerable caution.