First, you state your hypothesis that no relationship exists. Then try to disprove it using the appropriate comparison test. Your independent variable (nicotine) is categorical, which suits most tests, including ANOVA, but the dependent variable (levels of depression) is ordinal and not continuous or normally distributed, so you cannot use parametric test such as ANOVA. Therefore you should use non-parametric tests such as PERMANOVA, Chi Square, Mann Whitney or Kruskall-Wallis non-parametric one-way ANOVA analysis on ranks. If you obtain an overall p that is < 0.05, you can then do post hoc tests, such as the Dunn test :)

Dear Soffia,

First thing first, look for distribution of your data. This way, you can decide which test is applicable for your data (parametric or non-parametric).

You are talking about relationship. So, for your case you should go for correlation based tests (Pearson, Spearman. Based on what I've said and type of your data).

You have several problems. 1. Your DV appears to be ordinal with I believe 4 levels of depression. categories. This implies you need ordinal logistic regression because the variable is ordinal. and 2 your IV is categorical with 4 categories. BUT if DV is categorical with 4 categories you need multinomial logistic regression. Thus the important question for the moment is iif your DV has an order to it or not..see first screenshot.once you have decided if it's ordinal or multinomial SPSS has programs for both. Google search on SPSS and the name of the model and look at the videos to see how to use SPSS for this.

Here's a piece of advice. If you ever do something like this again. Plan it before before you collect data so you will be able to do this much easier. If you don't understand what you are doing at the beginning ask then. Don't wait until we all are is confused about your entire project. CALL A STATISTICIAN EARLY Before (to use a technical term) everything is screwed up. Ask again if you still need assistance but PROPER PRIOR PLANNING PREVENTS POOR PERFORMANCE. David Booth

You will probably want to treat both your variables as ordinal. If so, you could use the linear-by-linear test, which I believe SPSS calls "ordinal chi-square" and I think is included in the usual chi-square test of association output.

You could also use Kendall correlation as long as the procedure is good when there are many ties (Tau-b or Tau-c, I think).

Note, though, that neither of these tests treat the variables as dependent or independent. They test for a correlation. If you want to explicitly treat one variable as dependent, see the Cuzick or Jonckheere-Terpstra tests *.

However, it may be the case that you don't want to treat your independent variable as ordinal. In this case, a Kruskal-Wallis test would be an easy approach.

In any case, be sure to look at your data first. The correlation approach assumes that one increases (or decreases) as the other one increases. But it could be that the highest levels of your dependent variable are seen at a middle value of your independent variable.

Personally, I wouldn't assume one variable is independent and one is dependent. It could be that depressed people smoke more, rather than that nicotine addiction causes depression.

___________________________

* https://statistics.laerd.com/spss-tutorials/jonckheere-terpstra-test-using-spss-statistics.php

Hello Soffia Mz. Apologies if someone already asked this. Were the 4 ordered categories used at the time of data collection? Or did you collect more fine-grained data and carve the variables into categories after the fact? Thanks for clarifying.

A video on “Choosing the Correct Statistical Test (For Relationship between variables) with examples” https://youtu.be/1M5DQphChOg might be helpful.

I second Bruce Weaver 's request. I don't know of any depression measure that has just four categories. If you categorised a depression scale, be aware that this is a very bad idea. At best it loses you statistical power, and at worst it puts the breaks in the wrong places and you miss the pattern.

So please describe your methods a little more. Previous posters have pointed you to a number of ways of doing what you want to do, but Bruce – and I – are concerned that we don't understand why you want to do it.

That same pointless response has been posted to dozens (if not hundreds) of other threads:

https://www.researchgate.net/profile/Driss-Benattabou/answers

Has your account been hacked, or are you just a spammer trying to increase your RG score?

UPDATE (5-Oct-2021): Driss Benattabou's account is no longer active. The link given above now takes you to a page allowing you to search for users.

Driss Benattabou

Yes, recommended