0
$\begingroup$

I'm formulating a questionnaire about recycling behaviour for my school's research project and I want to use the questionnaire to do relationship analysis between different variables or factors that influence recycling behaviour. So in my questionnaire there are sections asking about attitude towards recycling, recycling behaviour, obstacles to recycling etc. But since different sections ask questions differently and therefore have different scales (e.g section 1: 5 point likert scale, section 2: 4 point likert scale, section 3: choose options from A to E) Can I do relationship analysis with questions that have different scales? I'm especially concerned about whether my analysis is valid when comparing likert scale quesitons with categorical questions (choose A-E options).

Here is a picture showing some of the questions from different sections of the questionnaire. I edited them together in this picture.enter image description here

$\endgroup$
1
$\begingroup$

The scales you have are ordinal, not interval, so you can certainly use ordinal statistics with no problem: i.e. daily > 2x > 1x > less than 1 > don't. For correlations, this would be Spearman correlations.

You could also do crosstabulations and test via Chi-square, and get a measure of the magnitude of the correlation with Cramer's V or a similar statistic.

If you want to compare two subsets (e.g. the opinion of men vs women) you could do this by using a common survey metric: comparing the % "top box" or % "top 2 box" (e.g. % strongly agree, or % strongly agree + % agree).

No problem with any of that. If you want to use interval statistics, though (e.g. t-tests) you will need to assume the points are equally spaced. So you could assume daily=5 > 2x=4 > 1x=3 > less than 1=2 > don't=1, but that's arbitrary. Why this instead of daily=8 > 2x=5 > 1x=4 > less than 1=2 > don't=0? Since this is arbitrary, I'd try to do your school project without making these arbitrary assumptions.

$\endgroup$
0
$\begingroup$

You can do relationship analysis with quantities that have different scales but for the results to be highly accurate you would have to normalize/standardize your data. Normalizing is converting all your analysis inputs to the same scale.

You can convert categorical values A-E to numerical values that make sense before your analysis.

Hope this helps.

$\endgroup$
1
  • $\begingroup$ Hmm...I see what you mean, but I think that doesn't apply to my case. In my case I don't think I have to normalize my data since the differences in scale don't affect the results. If for e.g I want to see the relationship between Q1. which is asking how often they recycle with Q2 which asks their attitude towards recycling, the scales don't have to match since the degree to which a person recycles does not always equal to the degree of their positive/negative attitude to the environment. So the scales I think can never match. But one can still look at the relationship. Hope that makes sense. $\endgroup$ Apr 19 '17 at 9:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.