Which method to use? ANOVA, one way regression, or else?

#1
Hi guys,

For my research I have an independent variable 'education', which varies from 1 (low) to 6 (high) and a dependent dummy variable which is either 0 (do not know) or 1 (do know). I want to see whether higher educated people are more likely to know (dummy=1) the organisation than lower educated people.

I tried to use ANOVA in SPSS, but it doesn't seem to be working since my dependent variable only has 2 values (0 or 1). I watched some videos on YouTube but I can't find an example where the dependent variable is a dummy.. Can someone please help me?

Thanks in advance!!
 

Karabiner

TS Contributor
#2
You can use know/don't know as grouping
variable and education as test variable, and
perform a Mann-Whitney U-test. This will
tell you whether those who know have a
significantly higher education than those
who don't know.

With kind regards

K.
 
#3
Thanks for your reply, Karabiner!
I have never used a Mann-Whitney U-test before (I'm a beginner ;p ). Could you explain a bit maybe?

Thanks in advance!
 
#4
Also, I far as I can find in my textbook, in the Mann Whitney U test, the test variable needs to be continuous. However, education is a categorical variable, since it can only take the value 1 (lowest ed.), 2 (low ed.) etc, up till 6. Can I still use this test, or do I need another test?

Hopefully someone can help me!
 
#5
Try to do a logit model = logistic regression model, with know/don't know as dependent variable and education as independent variable. (A very similar model is the "probit model", that you can try for fun.)

You can also do a Pearson chi-square test for the two-by-two table.

Also, as a guidance to make the problem easier to understand and more concrete, calculate the proportions of "know" in each of the two conditions "educated" and "not educated" (and complementary, the "don't know" in each of the two conditions "educated" and "not educated").

If you are familiar with it, you can do a significance test of the proportions "knows" in the educated group versus the proportion "knows" in the uneducated group.
 
#7
OH, now I notice that that I had not read the original post careful enough. I thought that the the independent variable education only had 2 levels. Now I notice that it has 6 levels. That means that it need to be inserted as a "factor variable". (Or possibly a "regression variable" if the logit model can be described as a being linear in the raw education variable.)

For my research I have an independent variable 'education', which varies from 1 (low) to 6 (high) and a dependent dummy variable which is either 0 (do not know) or 1 (do know).
The proportions can be calculated and the chi-squared test can be done. It can be good to plot the proportion "know's" versus the education level.

I must admit that I don't know, I am not sure, if it is good the make a Mann-Whitney test with the dependent variable as a grouping variable and the independent as the variable to do the ranking on. Is it good to condition on the dependent variable and let the independent variable be the response? I guess that a relationship, an association, can be established, but does it not reverse the causality?

(But maybe I have not been reading careful enough! :) )
 
#8
Thanks for your help guys! I decided to use a Chi-square test to be sure that it's oké to use two categorical variables. The assumption of 20% was violated, but I saw in a YouTube video that in that case you can look at the likelihood ratio, which was 0,000, which means that there is an association. Hope I did it the right way now :)