Technical Note: Recommendations for Assessing Unit Nonresponse Bias in Dyadic Focused Empirical Supply Chain Management Research

Abstract

The last decade has seen an increase in empirical supply chain management research with dyadic data. Such data structures can further complicate the assessment of nonresponse bias, which plays a key role in establishing the credibility of research results. A survey of 75 research articles with dyadic data, published in five empirically focused supply chain management academic journals, over the last decade, reveals a lack of agreement on methods used in the assessment for potential unit nonresponse bias. Of the various statistical tests found, only the Multivariate Analysis of Variance (MANOVA) approach allows for a single statistical test to be utilized in assessing for potential unit nonresponse bias via incorporation of the design structure of the dyadic data. We investigate the use of an effect size confidence interval coverage, of a MANOVA, to detect a meaningful difference between respondents and nonrespondents correctly. Our results show that with dyadic data, such meaningful differences can be detected with significantly smaller sample size requirements than traditional approaches such as t-tests or ANOVA. Recommendations are provided for setting up and executing a MANOVA to assess for potential unit nonresponse bias with dyadic data.

Career satisfaction in the public sector: Implications for a more sustainable and socially responsible human resource management

“…We tested non‐response bias by comparing respondents who participated in both surveys (Clottey & Benton, 2020). We found no difference between the two ( p > 0.05).…”

“…We tested non-response bias by comparing respondents who participated in surveys (Clottey & Benton, 2020).…”

We can work it out: A multilevel examination of relationships among group and individual technology workarounds, and performance

“…To test for possible non‐response bias, we used a procedure common in OM research (Clottey & Benton, 2013, 2020), a comparison between participating respondents and non‐participating respondents. Specifically, we compared the participating respondents ( n 1 = 264) and the non‐participating respondents ( n 2 = 366) on the gender and age to test for nonresponse bias.…”

“…We received 275 valid responses from 68 groups, yielding a 69.4% response rate at the individual level and a 64.9% response rate at the group level. Considering that we did not have information about the nonrespondents, we thus conducted a comparison between the early 25% respondents and late 25% respondents (Clottey & Benton, 2013, 2020). It is assumed that late respondents were most similar to non‐respondents because their responses required more stimuli and a longer time (Clottey & Benton, 2013, 2020).…”

“…Considering that we did not have information about the nonrespondents, we thus conducted a comparison between the early 25% respondents and late 25% respondents (Clottey & Benton, 2013, 2020). It is assumed that late respondents were most similar to non‐respondents because their responses required more stimuli and a longer time (Clottey & Benton, 2013, 2020). Following the recommendation of Clottey and Benton (2013), we selected two survey items (i.e., GTW4 and ITW2) at random from all the survey items to assess nonresponse bias with t tests.…”