Hello Basia,

In essence, it doesn't really matter either way, except that it might be easier for you to think about what your results mean if you score your pain scale one way rather than the other. So, I'd begin by thinking about how you want to write about your results, and adjust your pain scale if necessary.  Of course, you would need to make which way you did the coding clear for your readers.

At least the recoding process, if you do it, is quite easy in SPSS - if you are using that package.

Robt.

Pearson's correlation coefficient requires continuous data. If the responses are mostly 0 or 4, there could be a problem. I recommend to also do a chi square test as you have frequency data for both variables.

To answer your question I believe changing or recoding the Pain variable as you have defined wouldn’t have any significant effect on your outcome (Inference) BUT I believe what you should be really worried about is the Test statistics here. I do believe you should consider a chi square Test as there are some categorical data and continous data from your frequency data (from your) self-reported data.

Hello again, Basia,

I hope I did not misunderstand something when I posted my initial answer up above, but the contributions from Raid and Issah make me have second thoughts.  It seems they have assumed that you have some categorical data, whereas I had assumed that by the time you were ready to enter your data into Pearson's correlation analyses all of those data would be continuous - either ordinal or equal interval. In particular, with the pain variable I had assumed you had a number of items, each scored on a 5-point Likert-type scale. I now realise that this might not be the case, and that you might have assessed pain with a single item.

Even if that is the case, it might be worth putting some thought into the best statistic to use.  If any of your data are more ordinal than equal interval in nature (none of your data appear to be categorical to me), perhaps Spearman's rather than Pearson's correlation coefficients would be appropriate. They usually don't differ much from each other, particularly in terms of statistical significance, and using Spearman's might allow you to avoid criticism from some quarters.

Chi-squares might also be worth using, as Raid and Issah suggest, but, depending on how many people you have in your sample, you might have too few entries in some cells, in which case a Fisher exact test might be necessary.

I must confess that my inclination would be to do what you seem to intend when posting your question, namely to use correlations.  I have come across a lot of research in which correlations are based on variables that have only 5 data points, and, indeed, I've done it in a lot of my own research, so I think that correlations might be OK. Perhaps check that out if you are concerned about it. I won't pursue it here because I think it's off-topic given your original question.

Regarding that question, I'd go back to my original answer and recommend that you code your pain item(s) in the way that is going to be easiest to think and write about, and let your readers know what you've done.

All the best with your research,

Robt.

Hello yet again, Basia,

I have come across the following site that might be helpful if you are trying to decide whether your pain data (and perhaps some of your other data) should be regarded as categorical or ordinal/continuous:

http://www.theanalysisfactor.com/can-likert-scale-data-ever-be-continuous/

Again, I realise that this is off-topic given your original question, but it might be useful.

Robt.

Thank you greatly for all your input! I have read all answers and am now contemplating on which tests to utilize and decide if my data type is categorical or continuous in nature. Thank you again, and happy new year!