How to Analyze Likert Scale Data: The Complete Guide

Likert scales are everywhere in research — from customer satisfaction surveys ("Rate your experience from 1 to 5") to academic questionnaires measuring attitudes, perceptions, and behaviors. Yet analyzing Likert data correctly is one of the most debated topics in statistics. This guide clears up the confusion and gives you practical steps for your analysis.

Likert Item vs. Likert Scale: A Critical Distinction

This distinction drives every analysis decision:

Likert Item: A single question with ordered response options.

"I am satisfied with my job." (1 = Strongly Disagree, 2 = Disagree, 3 = Neutral, 4 = Agree, 5 = Strongly Agree)

Likert Scale: A composite score calculated by summing or averaging multiple Likert items that measure the same construct.

Job Satisfaction Scale = Average of items Q1 + Q2 + Q3 + Q4 + Q5

Why it matters:

• A single Likert item is ordinal — the distance between 1 and 2 may not equal the distance between 4 and 5
• A Likert scale (composite of multiple items) is typically treated as continuous — the Central Limit Theorem and measurement theory support this when you have 4+ items

This distinction determines which statistical tests you can use.

Descriptive Statistics for Likert Data

For Individual Likert Items (Ordinal)

Use:

• Median (not mean) as the measure of central tendency
• Interquartile range (IQR) as the measure of spread
• Frequency tables and bar charts showing the distribution of responses
• Mode — the most frequently selected response

In SPSS:

1. Analyze → Descriptive Statistics → Frequencies
2. Click Statistics and select Median, Mode, Quartiles
3. Click Charts and select Bar Chart

For Composite Likert Scales (Continuous)

Use:

• Mean and standard deviation
• Histograms to check distribution shape
• Skewness and kurtosis values

In SPSS:

1. First, compute the composite: Transform → Compute Variable
2. Formula: MEAN(Q1, Q2, Q3, Q4, Q5) — using MEAN handles missing data better than summing
3. Then: Analyze → Descriptive Statistics → Descriptives

Which Statistical Tests to Use

For Individual Likert Items

Since individual items are ordinal, use non-parametric tests:

Research Question	Test
Compare 2 independent groups	Mann-Whitney U test
Compare 2 related groups (pre/post)	Wilcoxon signed-rank test
Compare 3+ independent groups	Kruskal-Wallis H test
Compare 3+ related groups	Friedman test
Association between two ordinal variables	Spearman correlation
Association between two categorical variables	Chi-square test

For Composite Likert Scales

Composite scales with 4+ items can typically be analyzed with parametric tests:

Research Question	Test
Compare 2 independent groups	Independent samples t-test
Compare 2 related groups (pre/post)	Paired samples t-test
Compare 3+ independent groups	One-way ANOVA
Compare 3+ related groups	Repeated measures ANOVA
Relationship between two variables	Pearson correlation
Predict outcome from multiple variables	Multiple regression

Important caveat: Even with composite scales, check that the data is approximately normally distributed (especially with small samples). If it is heavily skewed, use non-parametric tests or note the limitation.

Step-by-Step Analysis Workflow

1. Create Composite Scales

If your survey has multiple scales, compute each one:

SPSS: Transform → Compute Variable
JobSat = MEAN(JS1, JS2, JS3, JS4, JS5)
OrgCommit = MEAN(OC1, OC2, OC3, OC4, OC5, OC6)

Remember to reverse-code negatively worded items first.

2. Run Reliability Analysis

Before computing composites, verify internal consistency:

1. Analyze → Scale → Reliability Analysis
2. Enter all items for each scale
3. Check that Cronbach's alpha ≥ .70
4. Review item-total correlations (all should be ≥ .30)

See our detailed guide on Cronbach's alpha reliability analysis.

3. Check Descriptive Statistics

For each composite scale:

• Mean and SD
• Skewness (should be between -2 and +2)
• Kurtosis (should be between -2 and +2)
• Histogram to visualize distribution

4. Check Assumptions

For parametric tests:

• Normality: Shapiro-Wilk test or visual inspection of histograms and Q-Q plots
• Homogeneity of variances: Levene's test (for t-tests and ANOVA)
• Linearity: Scatter plots (for correlation and regression)

5. Run Your Statistical Tests

Based on the decision table above, run the appropriate test using the composite scale scores.

Presenting Likert Results

For Individual Items

Present a frequency distribution table:

Response	n	%
Strongly Disagree	12	8.0
Disagree	23	15.3
Neutral	34	22.7
Agree	52	34.7
Strongly Agree	29	19.3

Use a diverging stacked bar chart for visual impact — this shows the distribution centered around the neutral point.

For Composite Scales

Report means and standard deviations:

Job satisfaction was measured using a 5-item scale (α = .84). Participants reported moderate-to-high job satisfaction (M = 3.72, SD = 0.89) on a 5-point scale.

The Debate: Parametric vs. Non-Parametric for Likert Data

This is one of the most debated topics in applied statistics:

Strict view: Likert data is ordinal. Always use non-parametric tests.

Pragmatic view (widely accepted): Composite Likert scales with 5+ items approximate interval-level measurement. Parametric tests (t-test, ANOVA, regression) are robust to mild violations of the interval assumption, especially with larger samples.

What most researchers actually do: Treat composite scales as continuous and use parametric tests. This is accepted in most journals and by most thesis committees, especially when you demonstrate acceptable reliability (α ≥ .70) and approximately normal distributions.

Our recommendation: Use parametric tests for composite scales, non-parametric for individual items. Always report reliability and distribution characteristics so readers can judge the appropriateness of your approach.

Common Mistakes

1. Computing means of single Likert items — The mean of "Strongly Disagree to Strongly Agree" is not meaningful for a single item. Use median and frequencies instead
2. Not reverse-coding before computing composites — This is the number one data error in survey research. It makes your scale scores meaningless
3. Running parametric tests on a single ordinal item — A t-test on a 5-point item with only 5 possible values is questionable. Use Mann-Whitney U
4. Ignoring reliability analysis — If your items do not reliably measure the same construct (α < .70), the composite score is meaningless
5. Treating the neutral point as meaningful — "Neither agree nor disagree" often means "I did not read the question" or "It depends." Consider whether a forced-choice scale (no neutral) would have been better
6. Using the sum instead of the mean — Sums are affected by missing data. If participant A answered 4 out of 5 items and participant B answered all 5, their sums are not comparable. Always use the mean

Reporting in APA

Participants' attitudes toward remote work were measured using a 7-item Likert scale (1 = Strongly Disagree to 5 = Strongly Agree; α = .82). Descriptive statistics indicated a positive attitude toward remote work (M = 3.89, SD = 0.76). An independent samples t-test revealed a significant difference in remote work attitudes between managers (M = 3.52, SD = 0.81) and non-managers (M = 4.12, SD = 0.68), t(148) = -4.87, p < .001, d = 0.80.

Need help with your survey data analysis? From Likert scale validation to hypothesis testing and APA reporting, we handle it all. Request your free quote.