Likert Scale statistical tests

[rock.freeak667]

I am collection results from a questionnaire (this one to be exact) which contains likert scales. However after I collect my data I am not sure what exact tests I should run, for example, if I run ANOVA what would I select for the dependent factor? Or what are some other helpful tests that I can run on the data?

[Jogilvie]

The type of statistical test(s) you run will depend on what hypotheses you are testing. I.e., what was the purpose of the survey in the first place?? Theory/research questions will necesarily drive what tests you run on the data - the data itself does not tell you what tests should be run.

So you need to ask yourself - what was i intending to measure/assess with these scales? However, i'm assuming from looking at the questionnaire that your question may concern the scales themselves - I.e., do they actually measure what they were designed to measure. You might want to be running a Factor Analysis here to see whether the scales 'stick' together.

j

[rock.freeak667]

Thanks for that. I am told that I need to do ANOVA on the data which I collect, however, with respect to the factor I need to select what exactly would it be?

[terzi]

Hello rock.freeak667

I would not recommend ANOVA for likert scale data. Some assumptions will never be met: models can get pretty annoying. Maybe, and just maybe if you use some totals as response you could obtain something useful and even that way I should use some non parametric alternative to ANOVA (Kruskal-Wallis test, Friedman test).

Anyway, after looking your survey, I don't really understand what you want to obtain with an ANOVA model. I think you could be interested in some other statistical procedures, such as Logit Models or Correspondence Analysis. Even Canonical Correlation seems plausible. Everything will depend on what kind of results you desire.

[rock.freeak667]

Anyway, after looking your survey, I don't really understand what you want to obtain with an ANOVA model. I think you could be interested in some other statistical procedures, such as Logit Models or Correspondence Analysis. Even Canonical Correlation seems plausible. Everything will depend on what kind of results you desire.

Well I am supposed to use that survey to measure knowledge management, but my supervisor just told me that I need to do more statistics tests other than the normal mean, standard deviation, variance,etc. stuff and do things like ANOVA on the results. Other than that, my research into likert scales and other surveys really did not show me any other tests to run.

[terzi]

If you want some interesting results, try with cluster analysis. That way you will be able to find groups of people that responded your questions similarly. You can get one group of people with no knowledge at all, another with average knowledge, another one with good basic knowledge but not advanced one, etc.

This analysis is very common for market segmentation, I think that could work for you.

Good luck

[rock.freeak667]

If you want some interesting results, try with cluster analysis. That way you will be able to find groups of people that responded your questions similarly. You can get one group of people with no knowledge at all, another with average knowledge, another one with good basic knowledge but not advanced one, etc.

This analysis is very common for market segmentation, I think that could work for you.

I think you could be interested in some other statistical procedures, such as Logit Models or Correspondence Analysis. Even Canonical Correlation seems plausible. Everything will depend on what kind of results you desire.

Thanks for the information. Are there any sites that can tell me how to interpret the results that come from these tests? I did also add in a question to indicate groups so I can see how knowledge is managed amongst the groups. I will do some additional research into the Logit model and the Correspondence analysis.

Everything will depend on what kind of results you desire.

I am not too sure as to what you mean by the kind of results (mostly because I am not sure how my results should be like ideally!).

[gianmarco]

Hi,
as for Correspondence Analysis, you can give a look to this long earlier discussion on the topic.

Hope this helps,
regards
Gm

[sg2kilo]

there is 4 section in questionnaire,
1. knowledge of management process
2. leadership in knowledge management
3. knowledge management culture
4. knowledge maangement technology.
All section consist of several question in likert scale.

I'm agree with Jogilvie, choosing the test is depend on your hipothesis.

Another consideration is the type of your data, is it nominal, ordinal, or ratio-interval.
When we talking about likert scale, the type of data is ordinal. So you can not simply run anova on your analysis, because anova consist of several assumption like:
- independent group testing for more than 2 group
- the data distribution is normal
- the variance of all group should be the same.

Is it possible to analyse your questionnaire with anova?
Yes we can... we only use the total score on each section and analyzing those section in different group. But first of all we must test the distribution of data using 1 sample kolmogorov smirnov test or shapiro wilk on each group.

[CowboyBear]

The type of statistical test(s) you run will depend on what hypotheses you are testing. I.e., what was the purpose of the survey in the first place?? Theory/research questions will necesarily drive what tests you run on the data - the data itself does not tell you what tests should be run.

This, this, a thousand times this!

[rock.freeak667]

Another consideration is the type of your data, is it nominal, ordinal, or ratio-interval.
When we talking about likert scale, the type of data is ordinal. So you can not simply run anova on your analysis, because anova consist of several assumption like:
- independent group testing for more than 2 group
- the data distribution is normal
- the variance of all group should be the same.

Is it possible to analyse your questionnaire with anova?
Yes we can... we only use the total score on each section and analyzing those section in different group. But first of all we must test the distribution of data using 1 sample kolmogorov smirnov test or shapiro wilk on each group.

I have another question which asks for the job position, but that would be ordinal data, but I don't get why the other questions are not scale type data.

For the hypotheses, guaging from the questions I don't suppose I can make them to be like :

"Job position 1 manages knowledge more effectively than job position 2"

Would I need to make hypotheses in general for each field of questions, e.g. one for leadership, one for culture etc.? I ask because at the end, I need to say whether or not knowledge is effectively managed amongst the sample group. (Sorry if I am sounding a bit uneducated, statistics was never my favorite type of math)