How to Analyze Likert Scale Data: Step by Step
To analyze Likert scale data, first decide whether you are looking at single Likert-type items or a multi-item Likert scale. Summarize single items with frequencies, percentages, the median, and the mode, and chart them as bars. Combine several items into a scale score only when they measure one idea, and then you can use means and standard deviations.
This guide explains that difference, how to code and clean responses, and how to analyze the data step by step with AI or in Excel. It also covers charts, group comparisons, a worked example, and the mistakes that lead to misleading results.
What Likert scale data is
Likert data comes from survey questions that ask people how much they agree, how satisfied they are, or how often something happens. A typical five-point item runs from Strongly disagree to Strongly agree.
The University of Arizona's assessment team draws a useful line between two kinds of data. "A Likert Scale is a multi-item psychometric assessment tool that measures a concept across a series of individual questions." By contrast, Likert-type items are "individual items measured using various types of response scales."
That difference matters more than any formula. Boone and Boone made the same point in a 2012 Journal of Extension article. They wrote that the two "require unique data analysis procedures, and as a result, misuses and/or mistakes often occur."
A workplace survey might report Likert-type items one by one: "The onboarding prepared me for my role," "I would recommend this product," and so on. A research study might combine several related items into one Likert scale. Knowing which one you have tells you which statistics to use.
Likert-type items vs. Likert scales
The University of Arizona page reproduces Boone and Boone's suggested procedures for each kind of data:
| Measure | Likert-type items | Likert scale scores |
|---|---|---|
| Central tendency | Median or mode | Mean |
| Variability | Frequencies | Standard deviation |
| Associations | Kendall tau B or C | Pearson's r |
| Other statistics | Chi-square | ANOVA, t-test, regression |
For single items, the page is direct. They "should be analyzed as separate items using sorting and grouping techniques like frequency distribution tables or visual aids like bar charts." It adds: "Calculating means and averages is not appropriate for this data type, although data can be analyzed using median or mode."
For multi-item scales, the rules relax. The same page says the aggregated score "can be treated as continuous data and assessed using all measures of central tendency, mode, median, or mean."
The practical rule is simple. If you report one question, report its distribution. If you report a score built from several questions, you can report its average.
Before you start: code and clean the responses
Survey exports may store answers as text. Analysis is easier when each answer has a number, so start by coding them.
- Assign codes in order. For a five-point agreement item, use 1 for Strongly disagree through 5 for Strongly agree.
- Keep the labels. Codes are for sorting and counting. Your report should still use the words people saw.
- Flip reversed items. If a question is worded negatively, reverse its codes before you combine it with other items.
- Handle blanks and "Not applicable." Leave them out of the counts, and report how many there were.
- Check the scale length. Do not mix five-point and seven-point items in one table.
Keep a short codebook with each item's wording, the codes, and any reversals. It makes the analysis repeatable next quarter.
How to analyze Likert scale data
The three steps below use Powerdrill Bloom, but the same approach works in any tool that reads your survey export. The Excel version follows in the next section.
Step 1: Upload the survey export and code the responses
Export the responses from your survey tool as a CSV or Excel file, with one row per respondent and one column per question. Upload the file and describe the scale in natural language, such as "Columns D to K use a five-point agreement scale."
Ask for the text answers to be coded from 1 to 5, and list any reversed items so they are flipped. Then ask for the number of valid answers per item, and how many were blank or "Not applicable." Check those counts against your survey tool's response total before going further.
Step 2: Summarize each item with frequencies, the median, and the mode
Ask for a frequency table for each item: the count and the percentage for every response option. Add the median and the mode for each item, which are the measures Boone and Boone suggest for single items.
Some teams also report a top-two-box figure, the share who chose Agree or Strongly agree. It is easy to read and fits targets like the University of Arizona example of "90% of students report being confident."
If some items form a scale, ask for a combined score per respondent, then the mean and standard deviation of that score. Keep the item-level tables beside it, so readers can see which questions drive the score.
Step 3: Chart the results and compare groups
Ask for a bar chart of response percentages for each item, ordered from the most to the least agreement. Our bar graph maker page shows the chart side in more detail.
Check the chart against the frequency table before you share it. The bars should add up to 100% for each item, and the labels should match the wording respondents saw.
Then compare groups that matter to your decision, such as departments, regions, or customer segments. Ask for the percentage distribution of each item by group, so groups of different sizes stay comparable. Finally, ask for a short written summary that names the items with the lowest agreement.
How to analyze Likert data in Excel
Excel handles Likert analysis well once the responses are coded. Assume an item's codes sit in B2:B201, one per respondent.
Count each response. Microsoft describes COUNTIF as a way to count cells that meet a criterion, with the syntax COUNTIF(range, criteria). Count each option with =COUNTIF(B2:B201,1) through =COUNTIF(B2:B201,5).
Turn counts into percentages. Divide each count by the number of valid answers, =COUNT(B2:B201). Format the results as percentages.
Find the median. Use =MEDIAN(B2:B201). Microsoft notes a detail that matters for Likert codes. "If there is an even number of numbers in the set, then MEDIAN calculates the average of the two numbers in the middle." A result such as 3.5 means the middle falls between two answers, so report both labels.
Find the mode. Read it from the frequency table: it is the option with the highest count. If two options tie, report both.
Calculate top-two-box. Use =(COUNTIF(B2:B201,4)+COUNTIF(B2:B201,5))/COUNT(B2:B201) for the share who agree or strongly agree.
A PivotTable can do the same counting for many items at once. Put the item in rows, the response in columns, and the count of responses in values. Then show values as a percentage of the row total.
A worked example
Here is an illustrative result for one five-point item from 200 employees. The item reads "The onboarding training prepared me for my role." The numbers are examples, not data from a real survey.
| Response | Code | Count | Percentage |
|---|---|---|---|
| Strongly disagree | 1 | 10 | 5% |
| Disagree | 2 | 20 | 10% |
| Neutral | 3 | 30 | 15% |
| Agree | 4 | 90 | 45% |
| Strongly agree | 5 | 50 | 25% |
From this table:
- Median: Agree. Ordered from 1 to 5, the 100th and 101st answers both fall in Agree. The first three options hold 60 answers, and Agree holds the next 90.
- Mode: Agree. It has the highest count, 90.
- Top-two-box: 70%. That is 45% Agree plus 25% Strongly agree.
- Bottom-two-box: 15%. That is 5% Strongly disagree plus 10% Disagree.
A clear written finding would be: "70% agreed the onboarding prepared them, and 15% disagreed." That sentence is more useful to a manager than "the average was 3.75," and it follows the frequency-based approach for single items.
How to chart Likert data
Bar charts suit single items, and they are one of the visual aids the University of Arizona page names. For many items at once, a stacked bar chart shows each item's full distribution in one row.
One layout comes recommended in the statistics literature. Heiberger and Robbins make the case in the Journal of Statistical Software. "We recommend diverging stacked bar charts as the primary graphical display technique for Likert and related scales." In that layout, disagreement extends to the left of a center line and agreement to the right.
To build a simple version in Excel, put the disagree percentages in as negative numbers and the agree percentages as positive numbers. Then insert a stacked bar chart and sort the items by total agreement. The neutral share can be split across the center line or shown in a separate column.
Label each bar with its percentage, or at least label the agree and disagree totals. That puts the two headline numbers on the chart itself.
Whichever chart you use, keep the response order the same across every item. Use one color family for disagreement and another for agreement, so readers can scan the chart quickly. Our guide to turning messy survey data into clear charts covers chart cleanup in more depth.
Comparing groups
Survey results can raise a comparison: this team versus that one, this year versus last year. For single Likert-type items, compare the percentage distributions side by side rather than the averages.
Present the comparison as percentages within each group, because groups are rarely the same size. A team of 12 and a team of 200 can only be compared fairly as shares.
If you need a statistical test, Boone and Boone's table lists chi-square for Likert-type items. For scale scores, it lists ANOVA, t-tests, and regression. Small groups make any test less reliable, so report the group sizes next to the results.
Here is an illustrative example. On the onboarding item, 72% of a 150-person sales team agreed, compared with 58% of a 50-person support team. The fair way to report it is "72% of sales (n = 150) and 58% of support (n = 50) agreed." Readers can then judge the gap with the group sizes in view, rather than from two bare percentages.
Reporting the results
The University of Arizona guidance gives a clear warning about averages. It calls it "not appropriate to assign numerical values to the response scale during the analysis phase." The finding it warns against is "on average, students are confident."
A clear report can include four parts for each item or theme:
- The wording of the question and the response options.
- The number of valid answers and how many were left blank.
- The distribution, as a table or a chart.
- One plain sentence with the top-two-box and bottom-two-box shares.
If you track the same item over time, keep the wording and the response options identical. A changed word or an added option breaks the comparison with earlier rounds, so note any change in the report.
For a full write-up, our guide to a survey results report covers structure and layout. If the findings are going into a meeting, the guide to a survey results presentation covers the slides.
Common mistakes to avoid
- Averaging single items. A mean of 3.4 hides whether people are split or lukewarm, and the guidance above advises against it.
- Mixing item and scale statistics. Use frequencies and medians for items, and means only for multi-item scale scores.
- Forgetting reversed items. A negatively worded item that is not flipped will pull a scale score the wrong way.
- Dropping the neutral option silently. If you exclude Neutral from top-two-box, say so in the report.
- Comparing raw counts across groups. Use percentages within each group instead.
- Over-reading small samples. A shift from 60% to 70% agreement means little if only 10 people answered.
When your responses arrive as a messy export, you can try Powerdrill Bloom to code them and build the frequency tables and charts.
Frequently asked questions
How do you analyze Likert scale data?
Code the responses as numbers, then summarize each item with frequencies, percentages, the median, and the mode. Chart the distributions as bars. If several items form one scale, combine them into a score and report its mean and standard deviation.
Can you take the average of Likert scale data?
For single Likert-type items, guidance based on Boone and Boone advises against means and recommends the median or mode. For a multi-item Likert scale score, the mean and standard deviation are appropriate.
What is the best chart for Likert scale data?
Bar charts work well for single items. For many items, a stacked bar chart shows each item's full distribution. Heiberger and Robbins recommend diverging stacked bar charts as the primary display for Likert scales.
Is Likert scale data ordinal or interval?
Single Likert-type items are ordered categories, which is why guidance based on Boone and Boone suggests the median and mode. Scores built from several items can be treated as continuous data, according to the University of Arizona page.
How do you analyze Likert scale data in Excel?
Use COUNTIF to count each response, divide by COUNT for percentages, and use MEDIAN for the middle response. A PivotTable can count many items at once. Then build a stacked bar chart from the percentages.
Sources: Boone and Boone, Analyzing Likert Data, Journal of Extension (2012) · University of Arizona, The Analysis of Scale Data · Heiberger and Robbins, Design of Diverging Stacked Bar Charts for Likert Scales, Journal of Statistical Software · Microsoft Support, COUNTIF function · Microsoft Support, MEDIAN function.