Ordinal or nominal level variables would generally be consider factors while scale variables would be covariates. The dialog boxes for Poisson have controls for both. In ordinary regression, you would have to convert a factor to a set of dummy variables.
Note: a dichotomous variable could be treated either way.
--
Original Message:
Sent: 1/12/2024 2:07:00 PM
From: Manou Prinsen
Subject: RE: Drowning in options
Dear Jon,
Thank you so much for your reply. I am going to look into Poisson and negative binominal regression, and redo my correlations with Spearman.
How do I differentiate between a simple linear term and a factor? Does that require a different type of test or is it something to keep in mind when analysing the data?
Again, thank you for the comprehensive reply. Have a great weekend.
------------------------------
Manou Prinsen
------------------------------
Original Message:
Sent: Fri January 12, 2024 01:48 PM
From: Jon Peck
Subject: Drowning in options
If your dependent variable is a count, consider Poisson or negative binomial regression. You can find this via Analyze > Generlized LInear Models > Generalized Linear Models. I suggest that you go through the case studies for this via Help > Topics even though time is short.
If you are looking at simple correlations, you could just use Spearman or Kendall.
Note that your ordinal predictor should probably be treated as a factor, not just a simple linear term.
--
Original Message:
Sent: 1/11/2024 9:17:00 PM
From: Manou Prinsen
Subject: Drowning in options
Dear all,
Thank you for opening my post. I am new on this blog so if anything is in the wrong place, please excuse me.
I am currently writing my bachelor thesis and I have a big deadline tomorrow. (It is currently way past midnight here).
Unfortunately, I just found out that the tests that I have run to see if there are relations between different variables, were not allowed to be done. I used Pearson's correlationcoefficient but as it turns out, my data is not normally distributed. Why I chose Pearson, I would not be able to tell you anymore.
So, after finding that the tests that I have run are not scientifically sound, I tried to find what test is actually the one that can answer my question without me ignoring the minimum requirements for the test. And now I am lost. I have not found my test yet and am beginning to become quite desperate. Below some information to make clear what I would like to test.
I did a survey with a little over 200 respondents. My research is about consumer understading of eco-labels, and so I asked my resondents to answer some personal questions (age, gender, location, education, and if they were working in the hair industry which is the target industry). Those variables are the independent variables, age = ratio, education = ordinal, and the other three are nominal if I am correct. And then asked them to answer 14 questions about eco-labels, all multiple-choice, the amount of correct answers is then my dependent variable, and a ratio variable.
Now I am curious to see if I can find any relation between the personal questions and the amount of correct answers, but I just cannot seem to find which test I should and am allowed to run. As mentioned before, the data is not normally distributed (makes sense as they are not continuous). Age for example is really skwered towards students and for location I have 90% in one location and 10% in the other.
If you have gotten this far, thank you for sticking. If you have any clue which test would help me answer my question, it would be greatly appreciated. I am going to bed, hopefully tomorrow will bring answers.
------------------------------
Manou Prinsen
------------------------------