I suspect your problem is due to indirect effects that take the form of what is known as "supression." This can happen when the direct effect of status inconsistency is in the opposite direct from its indirect effect via education.

But I think you have a bigger problem here, which is how you measured status inconsistency. I'm guessing that you want to look at is the "discrepancy" between education and income, which is unlikely to be correlated with eduction, since people with either high or low education can be either higher or lower than expected on income. Yet your current measure indicates that people with higher education are much more likely to have status inconsistency. Does that really match your theory?

As for the multicolinearity between your two dummies, it sounds like what you really want to do is test for the interaction between them, which is straightforward. First run a regression that contains only the two dummy variables, then run a second regression also includes the product of the two (i.e., the interaction effect). What you want to check is the change in R-sq between the two equations to find out if the interaction adds additional explanatory power, above and beyond the two main effects.

Hi, in the first correlation 0.593 is accepted on the average but how much is the Approx. Sig.?

How did you combine education and occupation to get the variable "statusinconstency"?

Would like to know about your model proposed to figure out the last variables mentioned.

REgards

Hey,

The significance of the variabeles in the regression?

I combines education and occupation through extracting one from the other, so I get a scale from -4 to 2 , where -4 are those who are most overrewarded (low education and high profession), and those at 2 are status inconsistent who have high education and low jobs. 

My model is the influence of precarity on political participation.

Political participation has items which are always looked at seperatly through binary logistic regression.

Main variables: income and job insecurity + interaction between both

Controls: age, gender, religion (and education but i excluded it...)

I suspect your problem is due to indirect effects that take the form of what is known as "supression." This can happen when the direct effect of status inconsistency is in the opposite direct from its indirect effect via education.

But I think you have a bigger problem here, which is how you measured status inconsistency. I'm guessing that you want to look at is the "discrepancy" between education and income, which is unlikely to be correlated with eduction, since people with either high or low education can be either higher or lower than expected on income. Yet your current measure indicates that people with higher education are much more likely to have status inconsistency. Does that really match your theory?

As for the multicolinearity between your two dummies, it sounds like what you really want to do is test for the interaction between them, which is straightforward. First run a regression that contains only the two dummy variables, then run a second regression also includes the product of the two (i.e., the interaction effect). What you want to check is the change in R-sq between the two equations to find out if the interaction adds additional explanatory power, above and beyond the two main effects.

It is actually based on a theory I read in a book, and the measure I took from another study.

The theory is what I am testing which says: those with high education who get lower jobs often will participate in protests. 

I could also divide it in 3 categorical groups. Those overrewarded, normal rewarded, and under rewarded

I have around 9 independent variables (all individual statements that show a form of political participation). I later in the discussion group them together to talk about the results (formal forms, activist forms...). But I want to see specifically where the difference is in participation.

Ok so I ran the regression first with only the 2 dummies and Nagelkerke R Square is .012 (so without any controls or anything).

Then I add the interaction between those and Nagelkerke R square is .019

This is only for one of the independent variables.

When I try another one Nagelkerke is first .040 and with interactions it is .046. 

block and model is not significant but one of the interactions is 

This is the correlation. opleidingsniveau = education

Statusinconsistency you see, written almost the same.

I made statusinconsistency into a categorical, so that I have 2 dummies : over- and underrewarded against the reference group 'normally rewarded).

The correlation of the dummy overrewarded has a correlation of abit more than 0.4 with educational level 

I was thinking, maybe i should take normally and overrewarded together as category 0, and underrewarded as category 1 ?

Because all im getting now with reference category 'normal', is an effect of the overrewarded, but not effect from those who are underrewarded. But there might be an effect of underrewarded when the reference are those who are overrewarded. 

However,  ill end up with a big group vs a smaller one

Right now I have:

59.5%  normal

22.3% overrewarded

18.2% underrewarded 

This may be due to multi-collinearity.  Run Klein's stepwise regression. If due to the additional variable in the regression model, the sign or significance of the previous independent variable changes in the bivariate model, there may be presence of multi-collinearity.

Further, What is your basis for considering 0 .70  as the threshold level?

If the calculated  value of r is 0.59, you may further want to check whether this value is significant or not.