Multiple Regression Analysis with Categorical Variables:SPSS ERRORS

[Sidra]

Hi all, I need some urgent advice on multiple regression in spss. Let me provide you with background information on my ordeal.

I'm running multiple regression analysis with few categorical and numeric sociodemographic independent variables, with dependent variable being average score of respondents on stress measured on likert-type scale. One of my categorical variables is marital status with two category response (1-single ,2-married/widowed/divorced). Only respondents who were married/widowed/divorced were asked to provide information on another variable "childbearing status" by responding to question "do you have any children?". The response categories included yes, no and N/A(for single respondents). That's where my conundrum started. I needed to exclude N/A category from multiple regression analysis to make "childbearing status" a dichotomous variable, since multiple regression accepts only dichotomous categorical variables. Creating dummy variables didn't seem to me an appropriate option, so I coded N/A as missing value and selected option "exclude cases pairwise" for missing values while running multiple regression.

Now, the problem I am having is this spss warning in the output "For models with dependent variable Perceived Stress Score, the following variables are constants or have missing correlations: Marital Status of Respondents. They will be deleted from the analysis". The regression model automatically excludes marital status from the analysis and I somewhat understand the reason behind it; collinearity or missing values. Now, apparently I seem to be left with only one choice to resolve this problem that is leaving "marital status" or "childbearing status" out of the analysis. Both variables are important to me, so I want to keep them both in the analysis.

 I have been beating my brains out to solve this issue for last one year. I ''ll really appreciate if somebody who is adept at multiple regression and familiar with countering such problems can suggest me an alternative solution.

Thank you
(anxiously waiting for guiding responses)

[Maguin, Eugene]

Other responders may have different ideas about this problem. I note in passing that childbearing status has nothing to do with marital status; however, that decision has been made. I'd do this: combine marital status and childbearing status into a new variable with values of Single, MWD-no children, MWD have children. So three categories and from those make the following two contrast variables. Contrast1: Single = -2, MWD-no children = 1, MWD have children = 1. Contrast2: Single = 0, MWD-no children = -1, MWD have children = 1. Use Contrast1 and Contrast2 in the regressions.
Gene Maguin



-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Sidra
Sent: Monday, October 17, 2016 1:37 PM
To: [hidden email]
Subject: Multiple Regression Analysis with Categorical Variables:SPSS ERRORS

Hi all, I need some urgent advice on multiple regression in spss. Let me provide you with background information on my ordeal.

I'm running multiple regression analysis with few categorical and continuous demographic independent variables, with dependent variable being average score of respondents on stress measured on likert-type scale. One of my categorical variables is marital status with two category response (1-single ,2-married/widowed/divorced). Only respondents who were married/widowed/divorced were asked to provide information on another variable "childbearing status" by responding to question "do you have any children?". The response categories included yes, no and N/A(for single respondents). That's where my conundrum started. I needed to exclude N/A category from multiple regression analysis to make "childbearing status" a dichotomous variable, since multiple regression accepts only dichotomous categorical variables. Creating dummy variables didn't seem to me an appropriate option, so I coded N/A as missing value and selected option "exclude cases pairwise" for missing values while running multiple regression.

Now, the problem I am having is this spss warning in the output "For models with dependent variable Perceived Stress Score, the following variables are constants or have missing correlations: Marital Status of Respondents. They will be deleted from the analysis". The regression model automatically excludes marital status from the analysis and I somewhat understand the reason behind it; collinearity or missing values. Now, apparently I seem to be left with only one choice to resolve this problem that is leaving "marital status" or "childbearing status" out of the analysis. Both variables are important to me, so I want to keep them both in the analysis.

 I have been beating my brains out to solve this issue for last one year. I ''ll really appreciate if somebody who is adept at multiple regression and familiar with countering such problems can suggest me an alternative solution.

Thank you
(anxiouslywaiting for guiding responses)






--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Multiple-Regression-Analysis-with-Categorical-Variables-SPSS-ERRORS-tp5733279.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Sidra]

Eugene thanks so much for such a prompt response to my post. Excuse my ignorance, I'm interested in knowing why you chose coding the two dummy variables using numbers -2, and 1 for contrast 1 and 0,-1,1 for contrast 2.. Why cannot we choose 0 and 1 for contrast 1 and numbers 0,1,2 for contrast 2? Is there any particular reason to it? Would regression output differ in these two different cases of coding? I also want to know what appropriate labels(variable names) would be for these two contrasts that are to be entered into the model?

[Sidra]

CONTENTS DELETED

The author has deleted this message.

[Sidra]

Eugene, after your reply I digged into coding techniques and came to know about contrast and effect coding schemes. As far as I could figure out, you have used contrast coding to make it possible for the two variables 'marital status' and 'childbearing status' to be included in the model. I'm still trying to get a grip on this new information that I didn't have before. Your coding scheme, however, is making much more sense now. Waiting for more responses to come in!