The collapse of Likert scales during data analysis is an acceptable practice in certain circumstances, provided that the implications are carefully considered, and a clear justification is provided. The following points merit particular attention:

Justifications for Collapsing Likert Scales

1. Simplification of Analysis

A limitation of this study is the relatively small sample size. If the sample size is relatively limited, collapsing categories can assist in circumventing potential issues associated with an insufficient number of responses within specific categories. This can, in turn, enhance the reliability and validity of statistical tests.

Ease of interpretation is a further factor to be considered. The reduction of scale complexity can facilitate the interpretation of results, particularly for those without expertise in the field. This process reduces the complexity of the data, thereby facilitating the identification of patterns.

2 The objective of enhancing statistical power:

The consolidation of categories can assist in reducing sparsity in the data, wherein specific response options may exhibit a paucity of selections. This is of particular importance in the context of chi-square tests and other analytical procedures, where the expected frequencies should not be too low. The combination of responses that are conceptually similar is a further means of reducing the complexity of the data. If certain response options are conceptually similar and combining them does not result in the loss of meaningful information, collapsing can be justified. For example, the categories "strongly agree" and "agree" can be combined into a single positive response category.

3 Theoretical or conceptual justifications may be provided for such decisions.

If the distinctions between certain categories are not theoretically important or clear, collapsing them can provide a more meaningful analysis. For instance, in some contexts, the difference between "agree" and "strongly agree" may not be substantively significant.

Collapsing categories can sometimes align better with the research questions or hypotheses being tested, especially when a dichotomous outcome (e.g., positive vs. negative) is more relevant.

Depending on the study question and setting, it may be okay to collapse Likert scales from five or four ratings to three or two ratings. To help you decide, here are some things to think about:

Statistical Power and Analysis: If your sample size is small, lowering the number of groups can help your statistical power by making the sample size bigger. This could help you find important changes or connections in your data.

Simplicity and Interpretation: Less info can be harder to understand and share, especially with people who aren't tech-savvy. It makes the research process easier and the reporting of results less complicated.

Asking Questions: If your original Likert scale wasn't clear or if people had trouble telling the difference between neighboring groups (for example, "slightly agree" and "agree"), collapsing can make things clearer and more reliable.

It's natural for collapsing groups to get less information from responders. Depending on the goals of your study, this loss of detail might be fine, but it could hide small changes in answers.

Consistency and Comparability: Make sure that combining groups doesn't make it harder to compare your data to other datasets or books where the original scale was used in a different way.

You might want to do a risk analysis before choosing to collapse Likert scale groups. To do this, you need to look at your data with both the original scales and the collapsed scales to see if the results you came to are still true. This method helps make sure that combining groups doesn't change your results or hide important differences in how people feel or act.

Only when theres dependency. If not You make experiments with 5

While collapsing Likert scale categories might seem convenient, it is not advisable under Item Response Theory due to the loss of latent information, potential distortion of data, and the availability of models designed to handle the complexity of multiple categories. Maintaining the original categories (provided that the survey is fully validated) ensures that the analysis fully leverages the richness of the data, leading to more accurate and reliable results.