A Likert scale records an attitude through ordered response categories, commonly spanning “strongly disagree” to “strongly agree.” It is ordinal data: the responses have a natural order, but their labels do not supply a consistent numeric scale. This makes correct column typing and attention to how the survey was collected essential before drawing conclusions. section 6.1.2 section 6.4.1
Use a Likert scale when the question is about degree of sentiment rather than a measurable quantity. Favorability and trustworthiness of AI tools, for example, can be recorded from unfavorable or distrustful through neutral to favorable or trusting. First inspect the response distribution, including missing answers, because a skipped nonmandatory question may mean “no answer given,” not an invalid observation. section 6.4.2 section 6.1.2
Prepare Likert responses for comparison by:
- Preserving their response order instead of accepting alphabetical ordering. section 6.4.2
- Choosing and documenting where a placeholder such as “No answer given” appears; placing it first is a practical ordering choice, not an attitude level. section 6.4.2
- Making the column explicitly categorical with an ordered pandas Categorical type. section 6.4.2
- Cross-referencing ordered responses with a relevant subgroup, then normalizing within groups when comparing response proportions. section 6.4.2
Comparing AI-tool sentiment
An AI-sentiment field can be ordered as “No answer given,” “Unsure,” “Very unfavorable,” “Unfavorable,” “Indifferent,” “Favorable,” and “Very favorable.” A crosstab can then compare this distribution among current users, nonusers, and aspiring users. Restricting a heatmap to groups relevant to the question—for example, users and aspiring users—keeps the analysis focused on attitudes that inform the decision at hand. section 6.4.2
survey["AISent"] = (
pd.Categorical(
survey["AISent"].fillna("No answer given"),
categories=['No answer given', 'Unsure',
'Very unfavorable', 'Unfavorable',
'Indifferent', 'Favorable', 'Very favorable'],
ordered=True)
)
survey["AISent"].value_counts(dropna=False).sort_index()
The categorical conversion makes missing responses explicit and fixes the display order from negative to positive sentiment rather than alphabetical order. section 6.4.2
Likert results describe reported attitudes, not necessarily behavior or objective truth. Survey participants may omit answers, answer untruthfully, or differ systematically from people who do not respond. Missingness may also reflect survey flow, such as follow-up questions unavailable to some respondents; interpret response patterns with those limitations in view. section 6.1.2
