I know I have been asking a lot of questions, but this is how I learn best.
This time I have a question about control variables.
I'm looking at income (X1) and employment contract (X2) + the interaction (X1*X2) of those on political participation (Y). As controls I have age, and gender.
BUT, I would like to include political powerlessness, since it often is a direct cause of lack of political participation. Do you think it would be appropriate to include this in my model? I found in the literature precarious groups often have a higher political powerlessness, which leads to less political participation sometimes.
Or will this make things too complicated because if I control for it, I'm just gonna see how much of the variance in Y it explains. I can't use it as an interaction.
Ann, what you are talking about is not a covariate, it is an intervening or explanatory measure. Here is the difference:
A covariate is a measure that is related to both your predictor and your outcome, and may be an alternate explanation. For example, if age is related to income and political participation, any relationship between income and participation might instead be coming from age differences. You need to control for age to see if it is income or age that is the actual explanatory factor at work. You covary for age. In this analysis, you have your IV and your covariates in your model, and see whether there is a significant relationship between IV and DV after controlling for the covariates.
An intervening or explanatory measure (EM) is something that is in a causal pathway from your predictor to your outcome. Your predictor relates to the explanatory measure, which in turn relates to your outcome. This is a basic path analysis. The full impact of your predictor to your outcome is the *combination* of its direct effect (the relationship of IV to DV after controlling for EM) and its indirect effect (the relationship of IV to EM times the relationship of EM to DV). In this analysis, you need to run two regressions, one with the EM as the dependent variable and one with the final outcome as the dependent variable.
Hi Ann,
If powerlessness is likely to explain variance in your DV, and if your N is not too small, i would suggest to include it. Of course, there are also considerations of economy of design to be taken into account. If powerlessness can only be assessed with a lengthy questionnaire and if this might reduce the willingness to participate, you would have to weigh the potential loss of participants against the added information that the powerlessness measure would provide.
Best, Gerd
The questionnaire I use is that of our government, SCV-survey. It has N:1500 (altho smaller after I remove pensioners and students N:1023 or something). So I think this is big enough? Only 4 missings on the political powerlessness scale.
Lets say political powerlessness explains part of the variance in Y. But when including this, the coefficient on for example income goes down. So that suddenly income explains less of the variance. How do I explain this in words?
Ann, what you are talking about is not a covariate, it is an intervening or explanatory measure. Here is the difference:
A covariate is a measure that is related to both your predictor and your outcome, and may be an alternate explanation. For example, if age is related to income and political participation, any relationship between income and participation might instead be coming from age differences. You need to control for age to see if it is income or age that is the actual explanatory factor at work. You covary for age. In this analysis, you have your IV and your covariates in your model, and see whether there is a significant relationship between IV and DV after controlling for the covariates.
An intervening or explanatory measure (EM) is something that is in a causal pathway from your predictor to your outcome. Your predictor relates to the explanatory measure, which in turn relates to your outcome. This is a basic path analysis. The full impact of your predictor to your outcome is the *combination* of its direct effect (the relationship of IV to DV after controlling for EM) and its indirect effect (the relationship of IV to EM times the relationship of EM to DV). In this analysis, you need to run two regressions, one with the EM as the dependent variable and one with the final outcome as the dependent variable.
You might want to control for ethnicity. While it may covary with political powerlessness, it is not perfectly correlated. The more controlled for, the lesser the chance that you have a spurious effect. However, you must have an explanation for including the control variable. There is some research that suggests that sexual orientation and religion may also influence. For example, homosexual individuals may be more likely to participate to change the political environment to get laws passed.
yeah I was already thinking about controlling for ethnicity, altho the survey itself is only for belgians, so people who have lived here for at least x amount of years and are belgians now. But it can still give different results.
So education I already have plenty of literature to include this as control. I will also see what I canfind about sexual orientation and religion.
Thanks for the help!
I am reading about religion linked to political participation and its not really straight forward...
Identifying with a religieus tradition was found to decrease participation (http://www.ukessays.com/dissertation/literature-review/influence-of-religion-on-political-participation.php ), but participation in church activities made it rise (http://onlinelibrary.wiley.com/doi/10.1111/j.1540-6237.2008.00533.x/abstract) . yet another studie found that believing in a god who is directly involved in the world, like evangalistic protestant, have decreased participation and vise versa. Religious belief was more significant dan particcipation in church activities(http://esciencenews.com/articles/2008/06/24/religious.beliefs.significantly.tied.national.political.participation ). And accordign to others catholics and islam folks participate less http://www.ukessays.com/dissertation/literature-review/influence-of-religion-on-political-participation.php.
Etc... Im just beginning to read about this, but so far its getting more complex. And I only have "religious belief" in the survey. I also have church attendance but I dont know if I should include it al?
I dont have the question wether they believe God is actively involved in the world.
Ann, Your other complication is mixing variables that are discrete with non-discrete. Political powerlessness is not a numerical variable, thus has to be defined operationally, so you can measure it. Perhaps that has been done in your project, or you may be using an accepted measurement tool (for e.g. PP can be on a scale from 1-10), however, there may be more, a lot more explanation and evidence/support needed for the measurement of this variable than for the other (discrete) that you are using, i,e income, employment, age, gender, education, if added.
I would probably control for both if I had both. You are not making a hypothesis about the direction or magnitude of the correlation, just that these have been shown to have an influence on political activity.
Education (of course). if possible as numerical variable (years of schooling) to not unnecessarily complicate the model
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis