If your outcome is ordinal (with 7 ordered categories), an ordered logistic regression would be the correct way to analyze it.

PS: Please don't talk about "parametric vs. non-parametric" in this context. There are a hell lot of useful parametric distribution models out there (see links below); the normal distribution model is NOT the only one, and models other than the normal distribution are NOT neccesarily non-parametric. Standard analysis procedures are based on the assumption that errors (residuals) are normal distributed. This assumption may or may not make sense in a particular case. If it does not make sense, the next step is to find a more sensible distributional model and use a method based on this more sensible assumption. If this is not possible, one should actually ask oneself if one understands the system (the data) well enough to understand the question and to understand the figures the statistical analysis would spit out. It's easy to calculate statistics that I don't understand and that I am unable to interpret correctly or sensibly. Doing such analysis is mindless - but unfortunately very common.

http://www.stat.rice.edu/~dobelman/textfiles/DistributionsHandbook.pdf

https://www.itl.nist.gov/div898/handbook/eda/section3/eda366.htm

http://personalpages.to.infn.it/~zaninett/pdf/statistical-distributions.pdf

https://threeplusone.com/pubs/FieldGuide.pdf

https://home.cse.ust.hk/~dekai/4211/lectures/Exciting_Guide_to_Probability_Distributions-part1.pdf

https://en.wikipedia.org/wiki/List_of_probability_distributions

Handbook of Statistical Distributions with Applications (ISBN 9781498741491)

Introduction to Probability (ISBN 978 1138369 917)

I think you are a bit confused. Take a look at the attached notes, I think they will help. Best, David Booth

David Eugene Booth Thanks much for the attachment, I will go through. Yeah, I am super confused as I am new to this advance analysis.

If your outcome is ordinal (with 7 ordered categories), an ordered logistic regression would be the correct way to analyze it.

PS: Please don't talk about "parametric vs. non-parametric" in this context. There are a hell lot of useful parametric distribution models out there (see links below); the normal distribution model is NOT the only one, and models other than the normal distribution are NOT neccesarily non-parametric. Standard analysis procedures are based on the assumption that errors (residuals) are normal distributed. This assumption may or may not make sense in a particular case. If it does not make sense, the next step is to find a more sensible distributional model and use a method based on this more sensible assumption. If this is not possible, one should actually ask oneself if one understands the system (the data) well enough to understand the question and to understand the figures the statistical analysis would spit out. It's easy to calculate statistics that I don't understand and that I am unable to interpret correctly or sensibly. Doing such analysis is mindless - but unfortunately very common.

http://www.stat.rice.edu/~dobelman/textfiles/DistributionsHandbook.pdf

https://www.itl.nist.gov/div898/handbook/eda/section3/eda366.htm

http://personalpages.to.infn.it/~zaninett/pdf/statistical-distributions.pdf

https://threeplusone.com/pubs/FieldGuide.pdf

https://home.cse.ust.hk/~dekai/4211/lectures/Exciting_Guide_to_Probability_Distributions-part1.pdf

https://en.wikipedia.org/wiki/List_of_probability_distributions

Handbook of Statistical Distributions with Applications (ISBN 9781498741491)

Introduction to Probability (ISBN 978 1138369 917)

I dont think normal or non parametric data can affect the type of regression used

i think it depends on either the outcome is continuous or categorical ( binominal)

if the outcome is continuous use linear regression

Doaa El Amrousy My outcome is continous and does not follow the normal distribution. Jochen Wilhelm I am using 4 point likert scale and I tried to dicotomise the variable but the data is skewed towards one direction of category outcome and I am just getting 4 outcomes in one category. The OR shows from 2-284 and I feel reporting of this wouldn't make any sense to readers.

If you measured on a 4 point scale, why on earth do you want to dichotomize the outcome?* Being more consequent, you could throw away all your data, and the analysis would be even simpler. And if you would not do any experiment next time, you would further save money and time. That was possibly a bit too sarcastic, it's not meant in any offending way. I hope it will wake up some minds and make them start thinking (and possibly scare off those who don't want to think).

Still, given your original scale is meaningful, ordered logistic regression is the way to go. That should have been clear even at the time when the study was planned (and the scale was established).

---

*Wasn't that scale carefully established? If not - why not, and what kind of information did you expect to get from some kind of unvalidated scale? But if it was, how can one dare to throw this away?

As suggested by Jochen Wilhelm look at ordinal regression in the notes I sent. By the way as Jochen notes only methods are parametric or non-parametic not data. You might also look at this link as a start in choosing a method: https://www.google.com/search?rlz=1C1CHBF_enUS874US874&ei=OD7cXcL1MJDj_Abd34Zw&q=regression+methods+by+dependent+variable+properties&oq=regression+methods+by+dependent+variable+properies&gs_l=psy-ab.1.0.33i160.1170606.1206344..1211920...4.2..4.220.9317.0j53j5......0....1..gws-wiz.....6..0i71j33i22i29i30j0i362i308i154i357j0i273j0j0i131j0i67j0i70i249j0i22i30j0i333..10001%3A0%2C154.W2ptchAAaNU

Best wishes, David Booth

Uday Narayan Yadav , this is a genuine call for help. What determines the type of regression to use (if you MUST use it) is the predictor variable to a large extent. In your case, an ordered logistic approach could be attempted, but you need to study how robust the output will be.

Hello Uday Narayan Yadav ,

You don't need to assume Normal distributions to do regression . Rather than relying on a test for normality of the residuals, try assessing the normality with rational judgment

If you follow steven's categories of nominal, ordinal, interval and ratio, your Dependent Variable is ordinal, and should be analyzed with some form of ordinal logistic regression.

But the most common type (by far) of ordinal logistic regression is the proportional hazards model, which assumes proportional hazards. That assumption might be violated, in which case, you might want to use multi-nominal logistic. Collapsing the variable will only very rarely be correct. It throws away information, and that’s rarely a good thing to do.

I am trying not to add to the confusion but you write " The OR shows from 2-284 and I feel reporting of this wouldn't make any sense to readers. " which suggests that the problem is about the predictor(s) and not the response.

It would help to see a simple descriptive analysis of the response and predictors, especially a suitable graph of each one showing the distribution. To be clear it is not a requirement that the predictors are normally distributed.

BTW predictors are your IVs.

not the variable you are predicting. Good luck

In my case, I am used Brief Illness Perception(BIP) Scale to measure the illness burden with 250 samples. The scale have five domains and for each domain I have calculated final score. I am interested to look at factors associated with poor BIP. As far I know, the outcome need to follow normal distribution which running linear regression(not for a logistic regression) but in my case it turned out to be non-normal.

Then I used binomial LR , by categorising the total score of each domain/outcome(with 75% quartile cutoff) , which didn't work either because I got only 5 -7 outcomes in each domain. The problem I mentioned above that my OR ranges from 2-284(or should I recategorise the predictor like age, gender)

Kelvyn Jones Thanks much for your kind work. Very soon, I will be send you the descriptives and few analysis docs to you .

I am grateful to you all. Thanks for your time.