# Thread: What test to use?

1. ## What test to use?

Hi,

I am required to perform some statistical analysis on some data. I am measuring user opinion (given by a 4-point Likert Scale) on two different systems.

I have assigned numbers to the scale, such that 1=strong disagree, 4=strongly agree.

Is it satisfactory to simply compare the two tables - ie the modal value of system 1 is x, the modal value of system 2 is y - therefore one is better than the other?

Or, does some kind of statistical test need to be performed in order to even out errors associated with the average?

Sorry if I haven't explained this well enough, I'll be happy to clarify if necessary.

2. Originally Posted by samhogy
Hi,

I am required to perform some statistical analysis on some data. I am measuring user opinion (given by a 4-point Likert Scale) on two different systems.

I have assigned numbers to the scale, such that 1=strong disagree, 4=strongly agree.

Is it satisfactory to simply compare the two tables - ie the modal value of system 1 is x, the modal value of system 2 is y - therefore one is better than the other?

Or, does some kind of statistical test need to be performed in order to even out errors associated with the average?

Sorry if I haven't explained this well enough, I'll be happy to clarify if necessary.
Since you are handling categorical data, you may want to set up a contingency table. Then just run a chi-squared test at some fixed alpha value (level of your test). That may quantify things.

Contingency table - Wikipedia, the free encyclopedia

Here is the statistic used,

[Pearson's chi-square test - Wikipedia, the free encyclopedia

I'm not sure if this is the best solution to use. Likert scales give ordinal data, wouldn't reducing this to nominal data cause a 'loss' of information?

I was originally taking the mean from both Likert scales, and then calculating standard deviation. I'm not doing that anymore, as I don't think it's very accurate to treat Likert scale results as intervals.

Instead, I'm using the mode as a rough guide to the overall result for each Likert scale, and I'm using pie charts to compare the actual results.

Is this a more satisfactory method?

4. I think, in order to provide contextual information, I should give a sample question from my questionnaire:

"It is easy to create a game"
a) by hand:
Strongly disagree [] [] [] [] Strongly agree

b) using system:
Strongly disagree [] [] [] [] Strongly agree

5. You may have seen this, but this diagram may help you decide. I am not entirely sure how to approach this, but none of tests recommended in the diagram are that difficult.

Choosing the Correct Statistical Test

6. Thanks very much. I was reading Human Computer Interaction by Dix Et. Al, which suggested the Mann-Whitney method. Your diagram just confirmed my choice.

Thanks very much for your input