The way to statistically control some variables in any research depends on the design, objectives, statistical technique (to be used), nature of variables and their levels of measurement etc. I am sharing some ideas for some types of statistical designs.

1. Correlation (bivariate): If one wants to know the correlation between two variables and to control a third (or a set of third) variables then one can use partial correlation.

2. Multiple regression (correlation): To control the effect of one or more variables in multiple regression analysis one way is to perform hierarchical regression. In the first step enter the variables to be controlled and in the next step the predictors whose effect has to be evaluated (after controlling some other variables). In SPSS, use the method "enter' which is the default and use blocks to enter the IVs.

Another way to examine the relationship between two or more variables after controlling the effect of some other variables is to work with residual scores - the score obtained after controlling the effect of third variables. For obtaining residual scores the regression approach may be followed.

3. Group comparison design (use of such techniques as ANOVA, MANOVA). In such condition one can use ANCOVA, MANCOVA with the variables to be controlled as covariate.

All the above analyses can be performed through SPSS and that too using the regular analysis menu and dialog boxes related to said statistics. No Special SPSS macro or scripts are required for any of the said analyses.

Hope this helps a

Dear Naeem

سلام علیکم

The question you asked is very complicated to answer and there is no unique way for all tests. The following can help though

1- For correlation you can use "partial correlation" command of the SPSS

Analysis--> Correlation-->Partial

put the variable you want ot control in the "controlling for" tab.

The results would be controled for the varibale you put in the "controlling tab"

2- For ANOVA, you can use the ANCOVA command

1. From the menu at the top of the screen click on: Analyze, then click on General

Linear Model, then on Univariate.

2. In the Dependent Variables box put your dependent variable

3. In the Fixed Factor box put your independent or grouping variable

4. In the Covariate box put your covariate

Hope this helps.

Ramin

Hi Naeem, on the one hand, everything you mentioned, regression, t-test, ANOVA, correlation are all part of a general linear model, and as mentioned above, it's possible to introduce a "control" variable into procedures for all of these. On the other hand, the question of whether you can really "control for" the effect of a variable is hardly settled. You haven't really explained what you are trying to do when you are applying these controls in your models, but, for example, the use of control variables when you have a quasi-experiment, that is a comparison of two groups when you don't have randomization to treatment and control conditons has been highly questioned. I don't have any sources handy on this issue, but I imagine that a good statistical resource site, such as UCLA, probably has a discussion. Bob

Thanks Mustafa Raziq, Ramin Sadeghi and Robert Brennan for your guidance.

In response to Robert Brennan question, I actually wanted to see the effect of distress responses on flood related rumination among flood affected individuals. I want to control the effect of ' flood related damages and losses' using multiple regression analysis.

What I would suggest in your regression is that you look carefully at how these variables relate in combination. At the least I would look at a model without your control variable in it, which means the effect of that variable should be part of the error, then add that variable and see how that affects the variables in the model and see how that compares to your prediction. I would also see how your control variable is related to the variable or variables of interest in the model, such as by checking correlations. If they are moderately or more correlated, there may be difficulties interpreting the results because of the collinearrity. Bob

Thanks

Let say you want check the effect of education on earning controlling or removing the effect of age, the procedure is as follows

1 regress income on age (income= a +b(age)) and save the residuals (income*)

2 regress education on age (edu= c+d(age)) and save the residuals (education*)

3 The find the correlation of these two residuals.

by regressing in first and 2nd step we actually net out the effect of age in explaining these variables so by using residuals we actually use that part of variables which has not any effect of age.

for further discussion you can see Greene's Econometric Analysis

Thanks Fawad Zafar

If we want to statistically control the effect of some variable In analyzing the moderation effect, we may enter the control variables first, then we enter other independent and moderator variables; and in the last step, the interaction product of independent variable and moderator was entered. In this way, we control the effect of control variables.

The way to statistically control some variables in any research depends on the design, objectives, statistical technique (to be used), nature of variables and their levels of measurement etc. I am sharing some ideas for some types of statistical designs.

1. Correlation (bivariate): If one wants to know the correlation between two variables and to control a third (or a set of third) variables then one can use partial correlation.

2. Multiple regression (correlation): To control the effect of one or more variables in multiple regression analysis one way is to perform hierarchical regression. In the first step enter the variables to be controlled and in the next step the predictors whose effect has to be evaluated (after controlling some other variables). In SPSS, use the method "enter' which is the default and use blocks to enter the IVs.

Another way to examine the relationship between two or more variables after controlling the effect of some other variables is to work with residual scores - the score obtained after controlling the effect of third variables. For obtaining residual scores the regression approach may be followed.

3. Group comparison design (use of such techniques as ANOVA, MANOVA). In such condition one can use ANCOVA, MANCOVA with the variables to be controlled as covariate.

All the above analyses can be performed through SPSS and that too using the regular analysis menu and dialog boxes related to said statistics. No Special SPSS macro or scripts are required for any of the said analyses.

Hope this helps a

Thanks Rakesh.

its helpful

This is the link for UCLA it's very useful as Mr Robert mentioned

http://www.ats.ucla.edu/stat/spss/faq/

Maybe this video will help you

http://www.youtube.com/watch?v=xgA8qY63dX0

So if I want to avoid potential effects of trait worry (which is not a variable of interest in our study) on the occurrence of paranoid delusions, should I make regressions and compare if our IV's are correlated with a trait worry measure? Whatever the score participants have on that questionnaire, is it the only way to control for that? Thanks!

Hi, 

My objective is to find if affect is related to beauty. I collected data before (called: pre-affect) applying the stimuli; after application of the stimuli, I collect another set of affective data (called: post-affective). I need to "normalize" the two sets of data (pre and post) before conducting a regression analysis. Kindly explain if "pre-affect" can be considered a covariate, and can it be included into the Regression (SPSS) along with the "post-affective" data. Any help would be appreciated, and where should I include each set of data in SPSS. My dependent variable - 3rd dataset is "beauty"

Thanks

These answers are all very useful. Thank you for taking your time to write them!

That's helpful thanks 

I think Rakesh Pandey has said it all, depending on what you are looking at you can control the effect of some variables while using PASW (SPSS). Correlation and linear regression will deal with the relationship between two variables, t-test, one way anova will deal with comparing two variables. They are also multiple comparison of variables, you can used Post-hos.

Can anyone recommend a reference for inclusion or to control for demographic variables in regression analysis please?

Thanks

Rashmi

thank you for all useful answers