Definition: A Likert scale is a survey-based research technique used to collect ratings for items related to a common subject. Each item may be analyzed separately or can be summed to create a group score. Because of this summed total, Likert scales are also called summative scales.
Why It's Important: The output of a Likert scale gives you a simple data points to benchmark, analyze, and use in statistical models.
A Likert scale has various terms associated with it - below is an explanation of each to help avoid confusion and give you a general understating of this question type is structured.
This refers to the number of answer choices. In the example above there are five (5) points; this is called a 5-point scale. A 5-point scale is most common. Sometimes researchers use 7-point or 9-point scales. The odd numbers give respondents the option to choose a middle number. If a researcher decides to remove the "neutral" option, that will become a 4-point scale.
If a researcher decides to remove the "neutral" option then that would become a 4-point scale.
This is the rating you want to receive for each Likert item. So in the example above example, the response anchors are "satisfaction," with a score of 5 being perfectly satisfied and a score of 1 being perfectly dissatisfied.
A Likert item is a single question within a Likert scale. "Report layout" is the first Likert item in the example above.
Ordinal data that is concerned with order and ranking of a set of values. Because Likert scales ask for opinions or ratings, it is by definition ordinal data. This is commonly confused with interval data.
Interval data is concerned with the distances between two points. Common examples of interval data are temperature and weight. A Likert scale may have an even difference between each answer that researchers can score, summarizes, and measure, but when numbers are assigned to the values, this is called a latent variable.
A Likert scale question evaluates items related to a single topic. Here are some of the most common uses of a Likert scale with an example of what can be measured:
Here are the steps to create a Likert Scale:
You can upload an unlimited amount of Likert items. However, to help reduce respondent fatigue, we recommend including no more than 10 Likert items.
The scale length is essential when conducting a study using a Likert Scale. When Rensis Likert developed the scale in 1932, a 7-point scale was initially used. This provided three points in either direction plus the option of neutral. Below are some of the common types of scales used.
A scale with more points is often thought to provide more precision between answers. While this may be true, a consideration often overlooked is the time it takes for a respondent to answer. Longer scales may discourage respondents from completing your survey due to fatigue. We recommend using a 5-point scale as this is easy for respondents to fill out on both desktop and mobile devices and strike a balance between longer Likert scales.
When a researcher wants to force a respondent to make a choice and not choose neutral, a 4-point scale can be used. Anything lower than a 4-point scale would lead to less precise data.
This is one of the most popular lengths for a Likert scale. This scale provides two points on each side plus a single neutral option.
This scale provides three points on each side plus a single neutral option. This option gives slightly more precision than the 5-point scale while keeping the survey length manageable for respondents. Therefore, we recommend not over a 7-point scale when conducting a Likert scale study.
To increase accuracy A 10-point Likert scale is used. The 10-point scale has no neutral option and forces respondents to pick an answer and not choose neutral. It has 5 points on each side.
The response anchors are the top labels in a Likert scale. There are only two response anchors in the example above: "Dissatisfied" and "Satisfied".
In the example above, The Likert scale is formatted efficiently for digital devices by labeling only the two endpoints. If you want to define what answer or feeling each number is related to, include a key above the scale.
Numerous scale categories can be used depending on the project. The above example is a "Satisfaction Scale". 1 would represent perfect dissatisfaction, while 5 would represent perfect satisfaction. Here are some of the most common Likert scale categories.
When doing customer or product research, a Likert scale won't show you the relative importance of items compared to one another. For example, if a hotel asks how important various amenities are, guests will likely rate all features as important; this drawback could be referred to as "List Order Bias."
For research projects like this, a MaxDiff question should be used to determine what amenities are important in relation to one another.
Respondents may avoid answering at the extremes or end-points of each item (e.g "Very Dissatisfied") to avoid being perceived as having extreme views. This is called central tendency bias.
Respondents may agree with statements as presented or mark every answer as satisfied not to appear negative. Agreeing with all statements is called acquiescence bias.
The order in which Likert scale items appear can directly affect how people answer each Likert item. Depending on the study you are running, this may need to be accounted for. You can randomize the order in which each Likert item appears if you want to reduce question order bias.
Respondents may avoid negative answer choices for the fear that their answers will be used against them. This fear of repercussions is a general form of response bias. This is common in employment surveys. You can make your survey anonymous to combat this pitfall. Anonymous surveys can be used to collect more honest feedback and reduce this bias.
Likert items can also be summed to create a total score for each respondent. Because of this summed total, Likert scales are sometimes referred to as summative scales.
The most common way of displaying the results to a Likert Question involves showing a data table with the number of times each answer was submitted.
Technically speaking, Likert data is ordinal, so calculating the mean is not statistically accurate, but a weighted can still give you quick insights. A weighted average can get an average score for each Likert item or for the group of Likert items. The example below shows that the report layout has a weighted average of 2.3, which is clearly the lowest scored item. The weighted average gives quick, actionable data without a deep statistical dive.
Since Likert data is ordinal, using the median is considered the best practice to measure the central tendency for each Likert item.
If you wanted to use Likert data to build predictive models, you would need to use a particular type of regression called ordinal logistic regression.
For example, you could have a customer survey that starts with a Net Promoter Score followed by a Likert scale asking to rate the satisfaction for items like price, customer support, etc. Then, the output of a regression model might give insights that every 1-point increase in price satisfaction is associated with an increased 80% odds that the Net Promoter Score would improve by 1-point.
Psychologist Rensis Likert developed the Likert scale in 1932 for an article in the "Archive of Psychology" titled "A Technique for the Measurement of Attitudes." He attempted to build a scale that made it easy to judge attitudes and opinions.
Previous to this, the Thurstone scale was used. This method required a lot of work and the use of judges. It starts by asking a test group of respondents to write down their feeling toward a subject. The responses are collected, and then a set of judges would rate each response from 1 to 11, with 1 being very unfavorable and 11 being very favorable. The responses are then filtered down to 10 or less. Each response now has a value that is the average from the judges. The final questionnaire is now given to collect responses, asking them to mark what answer choices they agree with. This time, scores are calculated using the judges' values.
Likert proved that while an attitude can take the form of many words, there can only be a specific range that feeling can be in, generally from one extreme to the other. Furthermore, Likert proved that his method gave similar results to the Thurstone scale but with less work. The results were confirmed in another paper by Likert in 1934.