Reading Time: 5 Minutes
After my last post about the shoddy rating scale survey I received from an online retailer I've received a lot of questions about the different types of rating scales that one can use in quantitative research. So I thought it would be helpful to dive a little deeper into the two types of research scales I alluded to in my last post.
If you're not familiar with rating scales, they're a common type of closed-ended survey question used to ask a respondent to assign value to something, such as an object or attribute. There are many different types of rating scales, but the two most common types you'll come across are ordinal and interval level scales.
The Ordinal Rating Scale
An ordinal scale presents the response input options to the participant as an ordered set of categories, but with the distinction that the "distance between the categories [is] unknown". This just means that the response options can be ordered and ranked (i.e. strongly agree is greater than somewhat agree), but there is no known quantifiable distance between the intervals.
A common example of an ordinal rating scale is what's known as the likert scale, which is named after it's inventor, psychologist Rensis Likert. This type of scale is usually presented with a statement, or set of statements, and the participant is asked to rate how much they agree or disagree. In execution, ordinal/likert questions usually look something like this:
A few things worth pointing out here. First, when using an ordinal scale the focus is on the labels (e.g. strongly agree, somewhat agree, etc) when it comes to assigning value to the object or attribute. The labels indicate the magnitude of difference between the response categories, so it's important to use clearly differentiated wording. This is why we often see the same adverbs like very, somewhat, strongly, slightly as these help provide clear and obvious differentiation between the response categories. Ordinal scales also tend to utilize shorter ranges, such as a 4 or 5 point scale.
It's also worth taking a moment to cover the design of a likert scale, as it's quite unique in it's own right. As I mentioned earlier, rating scales are used to allow the research participant to assign value to an object or attribute. But with ordinal scales such as likert, the participant is asked to rate their level of agreement with a statement about said object and/or attribute, rather than rating it directly. Here's a breakdown of what I mean:
Interestingly enough likert questions are, in a way, one of the few cases where it's acceptable to use loaded questions in research as the statement often presents a biased position toward the object/attribute at hand. The scale then allows the respondent to overcome that bias by indicating how much they agree or disagree with the original statement.
When to use it?
Ordinal scales can be useful when it comes to latitudinal studies where the researcher is interested in understanding an audiences' perception or opinion of something at a specific point in time. Ordinal scale data can also be easier to work with, as you get compositional data which is typically spread across just 4 or 5 categories (i.e. strongly agree, somewhat agree, etc).
The Interval Rating Scale
An interval scale is similar to ordinal in that the response options can be ordered and ranked. But the key difference here is that the response options are numeric, hence the distance between the intervals is quantifiable (i.e. 4 is one unit greater than 3). An important distinction with interval scales compared to ordinal is that the focus shifts from the labels to the numbers, as it's the numeric values that indicate the magnitude of difference. Interval scales also tend to utilize larger ranges, such as a 10 or 11 point scale. In execution interval scales usually look something like this:
In the example above you can see that the question text provides instructions that help the respondent qualify how to interpret the scale (e.g. rate on a scale of 1-10, 1 being very poor 10 being very good). Then the respondent is asked to rate the object or attribute directly. This is an important distinction, as ordinal scales such as likert ask the respondent to rate the object/attribute indirectly through a statement.
You can also see in the above example that you don't actually need to label each interval on the scale. Instead, you can simply label the opposing ends and let the participant interpret everything in between. In fact, I would go so far as to say you should never label each interval on an interval level rating scale, for a few reasons. First, it could distract and confuse the respondent as they'll have to try and interpret value in the context of both the labels and the numeric scale. Second, with as many as 11 points on an interval scale you'll probably have trouble finding clearly progressive and differentiated labels for every interval. So less is more effective when it comes to labelling interval rating scales.
When to use it?
Interval scales are often used to give the researcher more precision in their measurement. This is because the respondent has more room for interpreting 'value', which in turn gives the researcher a lot more data. The benefit here is that you have more flexibility for analyzing things like standard deviation and how much variability there is in your dataset. I also find that there are generally more possibilities with how you apply interval scale data. For example, you can analyze your results simply by calculating the mean, median or mode, or you can cluster the results into ranges (similar to how NPS uses ranges for detractors, passives and promotors,
Furthermore, interval scales can be better suited to latitudinal studies where you want to be able to track a persistent variable over time.
The ordinal and interval level rating scales are probably the most common research scales you'll encounter, but they're by no means the only types of scales you can use. Which scale you decide to apply and how will largely depend on your research objective (RO). So never choose a scale without considering your RO, as well as the impact the scale will have on how the respondents interpret the question and what kind of data it will give you.