What is a Likert Scale?
The Likert scale is a well-loved tool in the realm of survey research. Named after psychologist Rensis Likert, it measures attitudes or feelings towards a topic on a continuum, typically from one extreme to the other. The scale provides quantitative data about qualitative aspects, such as attitudes, satisfaction, agreement, or likelihood.
Likert scale questions tap into the respondent’s level of agreement or disagreement with a statement. These questions go beyond a simple yes/no, allowing more nuanced responses. They provide a platform for respondents to express the intensity of their feelings about a statement, making them perfect for surveys.
Use Likert scale questions when you’re aiming for more than just a binary answer. When you want to measure the degree of sentiment towards something, these questions shine. They’re perfect for understanding customer satisfaction, employee engagement, or agreement with a statement.
Usually, Likert scales use a 5-point system. However, some surveys use a 7-point scale.
For instance, if you’re studying people’s attitudes towards a new product, a question might be, “I would recommend this product to a friend.” The respondent would then select their level of agreement from the provided options.
Read on for a Likert scale survey example, examples of scale options, tips on how to write effective questions, pros and cons, and how to analyze these survey data.
Likert Scale Survey in Action: A Hands-On Example
Let’s put theory into practice with a real-world scenario. Imagine you’re a product manager for an e-commerce company, and you’ve just launched a new website design. You’re interested in gauging user experience, including the website’s ease of navigation.
You create a Likert scale survey that includes the following question:
“The new website design is easy to navigate.”
The 5-point Likert scale options are the following:
- Strongly disagree
- Disagree
- Neither agree nor disagree
- Agree
- Strongly agree
Let’s say you collect the following 100 responses:
- Strongly disagree: 10
- Disagree: 20
- Neither agree nor disagree: 15
- Agree: 40
- Strongly agree: 15
Your initial interpretation: 55% of users (Agree + Strongly agree) seem to find the new design easy to navigate, while 30% (Disagree + Strongly disagree) find it difficult. A considerable 15% are neutral.
The mode and median are both 4, pointing towards a positive experience. However, the frequencies reveal a significant portion of users struggling with navigation, including 10% who strongly disagree. Those are people who really dislike your new interface. It’s worth reaching out to those users!
5-point Likert Scale Examples
5-Point Likert scales can assess various attributes, including agreement, importance, frequency, quality, and likelihood. Below are examples of 5-point Likert scales for different characteristics.
Agreement | Strongly Agree | Agree | Neither | Disagree | Strongly Disagree |
Importance | Very Important | Important | So-So | Less Important | Not Important |
Frequency | Always | Often | Sometimes | Rarely | Never |
Quality | Very Good | Good | Average | Poor | Very Poor |
Likelihood | Very Likely | Likely | Neutral | Unlikely | Very Unlikely |
Crafting Likert Scale Questions: Some Tips
You can successfully use Likert scale surveys for many topics, but you must write effective questions. Poor questions can ruin your survey results. Here are some tips.
- Stay Neutral: Be careful not to bias the respondent. Your questions should not steer towards positive or negative answers.
- Keep it Simple: Make sure your statement is straightforward. Avoid jargon or technical terms.
- One Idea at a Time: Don’t mix multiple ideas into a single statement.
- Specificity Matters: Be precise about what you’re asking. If your question is too broad, you’ll receive vague answers.
Pros and Cons of Using a Likert Scale
Every research tool has strengths and limitations, and the Likert scale is no different. Let’s dive into its pros and cons.
Pros
- Easy to Understand and Use: The Likert scale is straightforward for respondents. They easily grasp the concept of expressing their level of agreement or disagreement with the provided statement.
- Captures Range of Feelings: Unlike binary yes/no questions, this approach records a spectrum of feelings, providing richer, more nuanced data.
- Quantifies Subjective Experiences: The Likert scale quantifies qualitative aspects, converting feelings and attitudes into measurable, analyzable data.
- Versatile: It is adaptable across a wide range of research fields, from market research to psychology.
Cons
- Limited Depth: Likert scales provide little insight into the reasons behind responses. They indicate how someone feels, but not why they feel that way.
- Subjectivity of Responses: The interpretation of scale points can vary between respondents. One person’s “strongly agree” might be another’s “somewhat agree.”
- Response Bias: Respondents may lean towards “socially acceptable” answers or display a tendency to avoid extreme responses (central tendency bias).
- Ordinal Nature: Though analysts often treat Likert scale data as interval data, it is technically ordinal. The intervals between scale points are not guaranteed to be equal, which can complicate statistical analysis. More on this issue next!
Understanding these pros and cons helps you decide when and how to use this data type in your survey research. Consider these aspects carefully and design your study accordingly.
Related post: Nominal, Ordinal, Interval and Ratio Scales
Analyzing Likert Scale Data
Finally, let’s talk about crunching the numbers. Unfortunately, this point is where we see some limitations.
Likert scale data are ordinal, meaning the values have a specific order, but the intervals between values are not necessarily equal. That’s probably a bigger deal than it sounds. Learn more about Ordinal Data.
Frequently, you’ll code Likert scale options using the numbers 1 to 5. Using numerical codes in this context is a bit deceptive by making your data seem more quantitative than it actually is. While you can rank the data, you don’t know the exact spacing between values. For instance, the distance between 1 and 2 might not be the same as between 2 and 3.
Consequently, you can’t perform certain mathematical operations on ordinal data, such as addition, subtraction, multiplication, and division. Therefore, the mean is an invalid statistic for Likert scale data. Some analysts add multiple ordinal scores and treat the sum as a continuous variable. Unfortunately, that’s not valid either.
These shortcomings might seem like an obstacle but remember that you’re trying to take qualitative phenomena and make them as quantifiable as possible. Likert scale data can get you part way there. But you won’t have perfectly quantifiable measurements when assessing things like agreement and satisfaction because they don’t exist. And that limits the analyses you can perform.
Analyses You Can Use
Fortunately, valid summary statistics and inferential procedures exist for ordinal data from a Likert scale.
Descriptive statistics like the mode, median, and frequency distributions work well for Likert scale data.
To make population inferences about two groups in your sample, non-parametric hypothesis tests like the Mann-Whitney U or Kruskal-Wallis test are your best bet. Statisticians designed these tests for ranked data.
For measures of association between Likert scale items, you have several choices. Spearman’s correlation is a standard option for ordinal data. You can also use Somers’ D, Kendall’s tau-b, Kendall’s tau-c, and Goodman and Kruskal’s gamma.
While you can’t use the standard analytical choice with Likert scale surveys, you have options.
Finally, don’t just focus on the numbers. Use bar charts and line graphs to spot trends and patterns easily.
With these tips and tricks, you can effectively use the Likert scale in your research. Happy surveying!
Learn more about Nonparametric vs Parametric Tests.
Yulia Fransiska Putri says
I have a struggle, I have made a Likert scale with a gradation scale of 1-6 using the criteria “always, generally, often, sometimes, rarely, never/rarely”. However, I am confused about how to make the percentage from 1%-100%. I have difficulty finding sources
Mark says
Dear Jim
Thank you for your comprehensive answer.
Kind regards,
Mark
Mark says
Hi Jim
Thank you for this helpful post. Some of the survey software packages allow for respondents to use a slider when completing their likert surveys. So, for example, a five-point survey scale could receive interim scores. If your example was recoded (e.g., on a 20 point scale), it could look like this:
Strongly disagree 1
Disagree 5
Neither agree nor disagree 10
Agree 15
Strongly Agree 20
And with the interim scoring, a participant could score anywhere between 1-20.
Okay, sorry, long wind up to the question.
Is there a statistical basis for 1-5 vs 1-20 vs 1-40 vs 1-100 etc? I would have thought up to some point, a larger band should add more precision. Beyond some point, the precision becomes false.
Thank you for your consideration.
Mark
Jim Frost says
Thank you for your insightful question! As youโll see from this long response, it touches on various important aspects of scale measurement: precision vs. practicality, and what the implications are from a statistical perspective. If it’s TL;DR, then skip to the final paragraph for my conclusion!
First, you’re absolutely right that increasing the number of points on a scale can add more precision. By expanding a scale, we get closer to treating a discrete variable as continuous. From a statistical standpoint, having more points on the scale allows us to better model subtle differences in responses, and it can also improve the accuracy of statistical tests and parameter estimates.
A guideline in statistics suggests that if you have 10 or more equally spaced values, you can reasonably treat a discrete variable as if it were continuous. This is especially useful when conducting parametric tests, which assume continuous data. Thus, moving from a 5-point to a 20-point scale, or even higher, would theoretically provide this advantage.
However, there are a few challenges and considerations:
Equal Spacing: While you can create scales with many divisions, you can’t always assume that the data values between these divisions are equally spaced, especially with ordinal variables. For instance, the psychological difference between a score of “1” and “2” might not be the same as between “19” and “20”. So, even if the numbers are equally spaced, the underlying sentiment or intensity might not be. That limits the type of calculations you can perform on the data. Read my article about Nominal, Ordinal, Interval, and Ratio Scales for more information.
In short, you might well still be working with ordinal data and its limitations despite having more than 10 values on the scale that numerically appear equally spaced but the underlying phenomenon might not be.
Practicality and Response Behavior: While you might provide respondents with a scale that has many potential responses (e.g., 1-20), there’s no guarantee they’ll use the full breadth of the scale. It’s possible that despite being given a 20-point scale, respondents might predominantly use values analogous to the original 5-point scale (e.g., using predominantly 1, 5, 10, 15, and 20 and not using the other in-between values) thus not truly satisfying the guideline of having 10 distinct values. If many respondents hover around the traditional 5 points, then the added precision becomes more theoretical than practical.
False Precision: As you’ve astutely noted, beyond a certain point, added precision can be misleading or give a false sense of accuracy. If a respondent gives a score of “67” instead of “65”, can we confidently say that there’s a meaningful difference between those two responses? I canโt provide a general rule on that. Youโll need to use your subject-area knowledge and hopefully some measurement systems analysis to get a better idea for a specific case.
While expanding a scale can add more precision and make certain statistical procedures more justifiable, it’s essential to balance this with the practical aspects of how respondents interact with scales and the real-world interpretability of the results. The ideal scale might vary depending on the context and the specific research objectives at hand.
In conclusion, I share your reservations about the inherent benefits of merely extending a scale. Specifically, I’m skeptical of simply converting a traditional 5-point Likert scale into a 20-point one, especially when it’s used for subjective evaluations like in your example. Forcing additional precision might not be effective in such contexts. However, there could be specific situations where a more detailed scale is beneficial, especially when assessing inherently quantitative phenomena.
Federico says
Another enlightening article! Thank you.
Categorical variables can be either nominal or ordinal (ordinal are harder to handle).
A variable containing Likert scale data is an ordinal categorical variable coded in numerical form. To use an ordinal variable as input to a statistical model (linear regression, etc.), the variable itself must first be converted into a meaningful numeric format.
5-point Likert data already is in numeric format (integers from 1 to 5). But is it “ok” to assume that this data is linear, i.e. a score=2 and a score=3 are separated by the same distance that score=3 and score=4 are? That is an assumption…Is there a better way to deal with ordinal numerical data like the Likert scale data when we want to use it as input to a statistical model? Thank you
Jim Frost says
Hi Federico,
Ordinal variables are distinct from categorical variables. Remember, with categorical variables, there is no distance between the categories, it just different types or groups. Like literary genre, college major, profession, gender, etc. There’s no inherent order to those categories.
Ordinal variables have a natural order, such as small, medium, large.
However, ordinal variables are distinct from continuous variables because while there is an order, the differences between values may or may not be equal.
So, to answer your question, no you can’t generally assume that the differences between values are equal. Consequently, using ordinal data limits the types of calculations you can perform. For instance, you can’t add or subtract values. For more information about this issue, read my post about Nominal, Ordinal, Interval and Ratio Scales. In this scheme, nominal = categorical data. I show how the use of different scales affect the calculations and analyses you can perform.
There are some cases where you might be able to assume that the distances between values are equal. Usually that would be when you specifically designed the values to equidistant. There is some debate about that however.
Using ordinal data definitely complicates statistical analyses. For an input in a model (independent variable), you’ll have to choose to enter it either as a continuous or categorical variable. Each approach has pros and cons. The correct choice depends on the nature of the data (which provides the best fit), the sample size, number of ordinal values, and the goals of the research. Too much to cover here but I write about the choice in my Regression Analysis book.