GEE with Binary Data
Locked
 
|
Hello All,
I am new to generalized estimating equations and I am wondering if anyone in the group can help me answer a couple of data setup questions. I have binary outcome data on the same subjects over a seven year period (data collected on a yearly basis). Some subjects were present at year 1 and participated in the survey through year 7. Other subjects started in year 3 and continued to year 7. Still others participated only in years 1 thru 5, but not in years 6 and 7. I am trying to conduct a series of binary logistic GEEs with the logit link and autoregressive correlations. In conducting GEE, is it expected that all subjects have records on all years of the survey? What are the recommendations for setting up your data in this type of situations? Any advice, reading material, and/or annotated example would be greatly appreciated. Best regards, Mesfin ===================== 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 |
|
|
Mesfin,
You are permitted to have missing data without losing the remaining the data collected from a subject. It's important that the data are missing completely at random or at least missing at random. You need to set up your data in long (a.k.a. vertical) format. Here's an example of how to set up the dataset [with a couple missing points]: Subject Time Y 1 1 0 1 2 1 1 3 1 1 4 1 5 1 1 6 0 1 7 0 2 1 1 2 2 2 3 0 2 4 1 2 5 0 2 6 0 2 7 0 . . . . N HTH, Ryan Mesfin Mulatu wrote: > > Hello All, > > I am new to generalized estimating equations and I am wondering if anyone > in the group can help me answer a couple of data setup questions. > > I have binary outcome data on the same subjects over a seven year period > (data collected on a yearly basis). Some subjects were present at year 1 > and participated in the survey through year 7. Other subjects started in > year 3 and continued to year 7. Still others participated only in years 1 > thru 5, but not in years 6 and 7. > > I am trying to conduct a series of binary logistic GEEs with the logit > link and autoregressive correlations. > > In conducting GEE, is it expected that all subjects have records on all > years of the survey? What are the recommendations for setting up your data > in this type of situations? > > Any advice, reading material, and/or annotated example would be greatly > appreciated. > > Best regards, > > Mesfin > > ===================== > 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 > > -- View this message in context: http://old.nabble.com/GEE-with-Binary-Data-tp27772346p27776187.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 |
|
|
In reply to this post by Mesfin Mulatu
In addition to Ryan's notes, have you looked at the online help? Help > Case Studies, then Advanced > Generalized Linear Models > Generalized Estimating Equations gives an example of repeated measures binary logistic regression using the wheeze_steubenville.sav dataset that ships with the product. These data are also analyzed in Hardin & Hilbe's Generalized Estimating Equations. Alex |
|
|
In reply to this post by Mesfin Mulatu
On Thu, 4 Mar 2010 09:18:41 -0600, Alex Reutter <[hidden email]>
wrote: >In addition to Ryan's notes, have you looked at the online help? Help > >Case Studies, then Advanced > Generalized Linear Models > Generalized >Estimating Equations gives an example of repeated measures binary logistic >regression using the wheeze_steubenville.sav dataset that ships with the >product. These data are also analyzed in Hardin & Hilbe's Generalized >Estimating Equations. > >Alex Thank you Ryan and Alex for your help. It is good to know that I can include all of my cases in the analysis. I guess I have to read a bit on what exactly "missing completely at random" is and if it applies to my case. I will also check out the online help for additional insight. Regards, Mesfin ===================== 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 |
|
|
Mesfin,
I'm sorry I didn't give you the code in my initial response. Assuming time is to be treated as a categorical variable, link=logit, distribution=binomial, and you want an autoregressive var-cov matrix, then the code below is a good starting point: GENLIN y (REFERENCE=LAST) BY time (ORDER=ASCENDING) /MODEL time INTERCEPT=YES DISTRIBUTION=BINOMIAL LINK=LOGIT /REPEATED SUBJECT=id WITHINSUBJECT=time SORT=YES CORRTYPE=AR(1) ADJUSTCORR=YES COVB=ROBUST /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION . Much can be said about the code but time is against me right now. There are many issues that come to mind. Feel free to write us back if you have specific questions. HTH, Ryan
|
spss v.18 and R version???
|
|
Just installed V 18 of SPSS and would like to install R -- must I
install version 2.8 of R as per the online documentation (and do I install 2.8.0 or 2.8.1?) or will SPSS take one of the more current releases? Dan Bibel Massachusetts State Police (no, I'm a civilian and I can't fix your tickets) ===================== 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 |
|
|
In reply to this post by Mesfin Mulatu
Thank you so much Ryan for the codes. I will certainly use it as a
strating point. Let me ask for clarification about the missing data issue. In my dataset, some cases took part in the study at a later point (say Year 3) and other dropped out early (year 4 or 5). So we have missing data on all variables for years 1 and 2 for the late starters, and complete missing for years 6 and 7 for the drop outs. Does GEE accomodate this type of missingness? Most discussions of missingness focus on missing data only on a specific variables in the presence of others. Thank you. Mesfin ===================== 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 |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |