Elsevier

Tourism Management

Volume 33, Issue 3, June 2012, Pages 713-716
Tourism Management

Research Note
Using dummy regression to explore asymmetric effects in tourist satisfaction: A cautionary note

https://doi.org/10.1016/j.tourman.2011.08.005Get rights and content

Abstract

This research note addresses the misuse of standardized weights as measures of effect in dummy regression, which is the most frequently used technique for assessing asymmetric effects in the formation of tourist satisfaction. Unlike in regular regressions, standardized weights have no straightforward interpretation in dummy regressions, but they only carry the risk of providing misleading implications in theory building and guiding managerial action. To empirically underpin the arguments put forward in this note, an illustrative case example is used that provides insight into the underlying statistical mechanisms that cause unstandardized and standardized weights to provide significantly different implications in dummy regressions. The findings of this note should help to prevent bad practice in future studies that make use of the technique in assessments of asymmetric effects in customer satisfaction and/or the three-factor structure of customer satisfaction. However, the points put forward hold for dummy regressions in general.

Introduction

Multiple regression analysis with dummy variables is a popular technique used for exploring asymmetries in the relationship between the performance of particular product/service attributes and overall customer satisfaction (OCS). As revealed by Table 1, half of available peer-reviewed studies using this technique were published in journals related to travel and tourism (i.e. seven out of 14), and even 11 out of 14 studies in journals with a primary focus on service industries.

A review of these studies reveals, however, a widespread misuse of dummy regression which is very likely to result in misleading implications in theory and model building and in guiding managerial action—i.e. the use of standardized weights as measures of effect instead of unstandardized ones. Five out of 14 studies do not mention which type of weight was used, whereas only Alegre and Garau (2011) and Mikulić and Prebežac (2011) explicitly report use of the unproblematic (unstandardized) type. This brief note thus aims to raise awareness about this problem and to explain why standardized weights should generally be avoided when analyzing variable effect sizes with dummy regression. In elaborating the problem, the methodological foundations of dummy regressions in asymmetry assessments are first briefly discussed, followed by an illustrative case example that empirically demonstrates the problem and its implications.

Section snippets

Asymmetry assessments with dummy regression

When using dummy regression to explore asymmetric effects in customer satisfaction (CS), data about attribute-level performance (AP) and OCS are needed. Typically, these are collected via CS surveys using direct rating or Likert-type scales. The data are then used to perform a multiple regression analysis with binary-coded ratings of AP as predictors and ratings of OCS as the dependent variable:OCS=b0+jJ(pjdp,j+rjdr,j)+εjJ,where b0 is the constant, pj the incremental change in OCS as a

The problem of standardized weights in dummy regression

Although dummy regressions are methodologically appropriate for assessing asymmetric effects as described in the previous section, interpretations of results as previously described are reliable (in fact valid), only if analyses are based on unstandardized regression weights. This is because binary dummy variables do not have any true numerical meaning like variables in regular regressions, why there is no meaningful straightforward interpretation of standardized regression weights in dummy

Case study

To empirically underpin the points put forward, data from a survey on CS with an airline ticket purchase service are used (n = 786; the used data are part of a larger survey). AP and OCS were measured with 7-point Likert scales. In coding the data for the dummy regression only lowest and highest AP-levels were binary transformed into dummy variables (i.e. AP-scores of 1 and 7). Table 2 provides an overview of distributional characteristics of unstandardized binary dummy variables (i.e. number of

Conclusion

Although the statistics literature explicitly argues against the use of standardized dummy regression weights as measures of effect, there appears to be a lack of awareness about this issue in customer satisfaction studies that apply the technique. Neither is there a sound reason for preferring standardized to unstandardized weights in regular dummy regressions, nor can standardized weights be meaningfully interpreted. In fact, their use as measures of effect carries only the risk of providing

References (22)

  • J. Fox

    Applied regression analysis, linear models, and related methods

    (1997)
  • Cited by (35)

    • Application of ecological and aesthetic parameters for riparian quality assessment of a small tropical river in eastern India

      2020, Ecological Indicators
      Citation Excerpt :

      The 'Good' category, as defined in Section 3.6.3, was assigned a dummy value of 1, while the 'Poor' category was assigned a dummy value of 2 (Supplementary Table S3). The dummy weights also helped denote the average change in the dependent variable as a result of the independent variable assuming a particular value rather than a chosen reference value (Mikulic and Prebezac, 2012). Finally, the relationship between the perceived naturalness and attractiveness factors was examined through simple linear regression analysis.

    • Analysis of hotel services by their symmetric and asymmetric effects on overall customer satisfaction: A comparison of market segments

      2019, International Journal of Hospitality Management
      Citation Excerpt :

      This theory has been examined in many research areas, such as banking (Johnston, 1995; Matzler et al., 2007; Arbore and Busacca, 2009), health (Mittal and Baldasare, 1996; Matzler and Sauerwein, 2002), the automotive sector (Mittal et al., 1998), human resources (Matzler and Renzl, 2007; Matzler et al., 2004) and business-to-business relationships (Falk et al., 2010; Busacca and Padula, 2005; Šerić et al., 2016; Füller and Matzler, 2008; Ting and Chen, 2002). In the field of tourism, plane ticket sales services (Mikulić and Prebežac, 2012), animation programs (Mikulić and Prebežac, 2011a), congress services (Lee and Min, 2013), food-drink services at restaurants (Chen, 2012), website contents (Zhang and Cole, 2016; Kocabulut and Albayrak, 2017) and destination attributes (Alegre and Garau, 2011; Albayrak and Caber, 2013b; Mikulić et al., 2015) have been examined in the context of asymmetric relationships. Previous studies related to international hotel management have also addressed the asymmetrical relationships between service attributes and overall customer satisfaction.

    • Sources of satisfaction with luxury hotels for new, repeat, and frequent travelers: A PLS impact-asymmetry analysis

      2017, Tourism Management
      Citation Excerpt :

      In order to compare the structure of service quality factors for three market segments, a triple sample size as a single study is required. Furthermore, technically, extant asymmetry assessments with dummy regression are not easy to implement, so there are only a few applications in tourism studies as was mentioned in Mikulić and Prebežac’s (2012) study. In order to address the issues of large sample size and ineffective setting, a new asymmetrical analysis method is proposed with detailed instructions for researchers to follow.

    • Use of impact-range performance and asymmetry analyses to improve OTA website quality

      2016, Journal of Hospitality and Tourism Management
      Citation Excerpt :

      Fourth, the selection of the cut-off points for classifying different attributes may vary from the approach of other researchers although this condition is common in prior research. Fifth, the use of standardized coefficients of regressions was criticized by Mikulić and Prebežac (2012) despite being commonly used in most recent studies. However, their suggestion to use unstandardized coefficients was pre-mature and did not address the issues on cut-off points.

    View all citing articles on Scopus
    View full text