Missing values in hierarchical logistic regression

[Nogitsune]

Good afternoon,

I am trying to understand and perform a hierarchical logistic regression. Some of my variables are not dichotomous, so I am recoding them into dummy variables. There are some variables have multiple levels, two of which I need for the analysis. If I simply recode the variables into new ones, I will end up with quite a bit of missing data. My question is this: does the missing data affect logistic regression? I cannot use the multinomial option because it does not have the blocks for hierarchical analysis. I can either recode the two levels into 2 different dichotomous variables or recode into one variable and define the missing value. What would you suggest?

Thank you.

--

===================== 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

[Maguin, Eugene]

This is confusing because some explanation is missing and is needed.

>> Some of my variables are not dichotomous,

Are these  your independent variables (IVs) or your dependent variables (DVs)?

If they are IVs, there is no requirement that they be dichotomized and very good reasons for not dichotomizing.

 

>>My question is this: does the missing data affect logistic regression?

Yes. Absolutely. Missing data means a reduced sample size and that means reduced power, a coefficient has to be larger to attain a given level of significance, and possible bias in the results.

 

How about backing up a bit and explaining your analysis and where you see the problems being.

 

Gene Maguin

 

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Kseniya Katsman
Sent: Thursday, March 16, 2017 1:08 PM
To: [hidden email]
Subject: Missing values in hierarchical logistic regression

 

Good afternoon,

 

I am trying to understand and perform a hierarchical logistic regression. Some of my variables are not dichotomous, so I am recoding them into dummy variables. There are some variables have multiple levels, two of which I need for the analysis. If I simply recode the variables into new ones, I will end up with quite a bit of missing data. My question is this: does the missing data affect logistic regression? I cannot use the multinomial option because it does not have the blocks for hierarchical analysis. I can either recode the two levels into 2 different dichotomous variables or recode into one variable and define the missing value. What would you suggest?

 

Thank you.

--

Best,

Kseniya.

===================== 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

[Rich Ulrich]

"MISSING" will be left out of the analysis, so you don't want that.

You have your choice of how you create your dummy variables. 

For instance, if you have Catholic/Protestant/Other, you might create

    DUM1= Cath/all other, DUM2= Prot/all other.

That accounts for your available 2 degrees of freedom. 

On the other hand, if you want to account for the Cath/Prot as extremes,

while controlling for Other, you could take the artificial "linear trend" of

Cath= -1, Other=0, Prot=1  for DUM1, and DUM2= Other vs Cath or Prot. 

 - that will be easy to interpret so long as Other has trivial effect; but

you should not look at an equation without both DUM1 and DUM2 if

"Other" has much influence.

--

Rich Ulrich

---

From: SPSSX(r) Discussion <[hidden email]> on behalf of Kseniya Katsman <[hidden email]>
Sent: Thursday, March 16, 2017 1:07:32 PM
To: [hidden email]
Subject: Missing values in hierarchical logistic regression

 

Good afternoon,

I am trying to understand and perform a hierarchical logistic regression. Some of my variables are not dichotomous, so I am recoding them into dummy variables. There are some variables have multiple levels, two of which I need for the analysis. If I simply recode the variables into new ones, I will end up with quite a bit of missing data. My question is this: does the missing data affect logistic regression? I cannot use the multinomial option because it does not have the blocks for hierarchical analysis. I can either recode the two levels into 2 different dichotomous variables or recode into one variable and define the missing value. What would you suggest?

Thank you.

--

===================== 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