# How to use Likert Scale data in regression analysis?

#### irtiza104

##### New Member
My knowledge about statistics is elementary and I would really appreciate some help or suggestions in solving my current problem. I am doing a dissertation and I will collect the data using a likert scale. My question is can I use that likert scale data to run a regression analysis to find out the relation between dependent and independent variables?

For example,

Q1. I just need a clean bed and a decent bathroom in my hotel > 1 2 3 4 5 (Suppose, Factor 1)
Q2. I must have a direct bus service > 1 2 3 4 5 (Factor 2)
Q3. .... (Factor 3)

Suppose, I already know income and expenditure. Now how can I achieve this:

Expenditure = a + B1(income) + B2(factor 1) + B3(factor 2) + B4(factor 3)

Is there a way to do this? Or should I use a different approach?

I am actually trying to find relation between "domestic tourism expenditure" and different factors.

#### kiton

##### Member
Hello!

Assuming expenditure is a continuous variable, OLS estimation of your model is possible ("likert" regressors are not a problem at all). Yet, keep in mind, that there are many assumptions behind such estimation and those will require additional statistical explorations from your side.

#### Miner

##### TS Contributor
The thing to keep in mind about using Likert predictors, is that while your results will give you a fairly good handle on significance, directionality and relative contribution, it may not be terribly precise about predictions.

#### irtiza104

##### New Member
Hi, thank you both for your answers. Could you please mention how can i convert the likert scale data into numeric data so that I can use it on linear regression analysis? Should I just use 1-5 for strongly disagree to stonrgly agree and calculate?

Thanks

#### Miner

##### TS Contributor
You cannot convert it. Measurement scales are not equal. For example:
Ratio scale (zero has meaning): You can add, subtract, multiply and divide
Interval scale (zero has no meaning): You can add and subtract
Ordinal scale (e.g., Likert scale): You can rank

Ordinal scales such as a Likert scale only tell you that a rating of 3 is larger than a rating of 2. An interval scale would tell you that a 3 is 1 unit larger than 2. This is more information than provided by the Likert scale. A ratio scale would tell you that 3 is 1 unit larger than 2, AND that 3 is 1.5 times larger than 2. In addition, Likert scales are integer scales while interval and ratio scales may be subdivided to the limit of the measurement device.

In other words, a Likert scale simply does not provide the same information content as the other two scales. It's not a matter of converting.

When you use Likert ratings in regression, the results assume that the differences between a 1 and 2 are the same as the differences between any other adjacent ratings. In most cases this is simply not true. Therefore, used carefully, regression may provide insight into relationships, directionality and contribution. However, there will be higher than normal risks associated with your conclusions, and the regression model should not be used for predictions.

#### irtiza104

##### New Member
Okay, in that case, which model should I use? I want to find out the expenditure pattern based on customers perception or attitude (likert data)

I know that I maybe asking too much, but I have hit a wall.

Thanks

#### Lazar

##### Phineas Packard
Are you using individual items as predictors or are you wanting to calculate some kind of scale score? Also which stats package are you using?

#### irtiza104

##### New Member
I am planning to ask multiple questions regarding a predictor (like perception about transportation methods) and then combine them to get a score for "transportation" as a whole. I am using spss.

#### irtiza104

##### New Member
I have created a partial (not complete) questionnaire for the study. It might help to clarify my situation or what I am doing wrong. I have banded together the likert scale questions which I would like to calculate together. Attaching the questionnaire.