[Traminer-users] quantitative explanatory variables?

Matthias Studer Matthias.Studer at unige.ch
Sat Nov 5 17:22:36 CET 2011


Dear Juan Zuluaga,

Sorry, for the delay. Ok, I didn't understood that your question was 
more theoretical. I do not have a good answer to your question. We did 
not have an in-depth look at these methods or others that allow to 
include directly a quantitative covariate. I haven't thought about it 
before but the Mantel test may be meaningful in this context. The main 
problem I would see is about how to interpret the results.

All the best,
Matthias Studer






Le 05.11.2011 00:25, Zuluaga, Juan a écrit :
>
> Say, would a Mantel (1967) test work? The quantitative covariate can 
> be turned into a distance matrix.
>
> *From:*Zuluaga, Juan
> *Sent:* Monday, October 31, 2011 9:16 AM
> *To:* Users questions
> *Subject:* RE: [Traminer-users] quantitative explanatory variables?
>
> Mr. Studer, thank you very much for the code, it makes sense.
>
> However, are you implying that this is an open question?
>
> The fact that you are seem to be satisfied with MJ Anderson approach 
> for categorical explanatory variables and have implement it in 
> dissassoc(), while you have no equivalent routine for quantitative, 
> does is it mean that you are not satisfied with existing approaches 
> for quantitative explanatory variables?  May I ask you what have you 
> considered (and perhaps rejected)?
>
> -j
>
> *From:*traminer-users-bounces at r-forge.wu-wien.ac.at 
> [mailto:traminer-users-bounces at r-forge.wu-wien.ac.at] *On Behalf Of 
> *Matthias Studer
> *Sent:* Monday, October 31, 2011 2:57 AM
> *To:* Users questions
> *Subject:* Re: [Traminer-users] quantitative explanatory variables?
>
> Dear Juan Zuluaga,
>
> I agree with you. Our example dataset lacks an example with a 
> quantitative covariate.
>
> There are two solutions to analyse the link with a quantitative 
> covariate. The first one is to discretize the variable before using it 
> (an example is given below). The second solution is to use the tree 
> procedure. This procedure automatically finds the best cutting points 
> by testing all possible binary splits.  This will also work with 
> ordinal covariates.
>
> An example of both solutions is given below using the biofam dataset 
> (Swiss family life sequences between 15 and 30 years old).
>
> ## Loading TraMineR
> library(TraMineR)
> ## Loading the biofam dataset
> data(biofam)
>
> ## States labels
> bf.labels <- c("Parent", "Left", "Married", "Left/Married",  "Child",
>                 "Left/Child", "Left/Married/Child", "Divorced")
> ## States short labels for the sequences
> bf.shortlab <- c("P","L","M","LM","C","LC", "LMC", "D")
> ## Building the sequence object
> biofam.seq <- seqdef(biofam[,10:25], states=bf.shortlab, labels=bf.labels)
> ## Computing distance using Optimal matching with transition based 
> substitution costs.
> biodist <- seqdist(biofam.seq, method="OM", sm="TRATE", indel=1)
>
> ## First solution : Use a discretized variable
> ## The "cut" function creates a factor using the given cutting points
> biofam$cohort <- cut(biofam$birthyr, c(1900, 1930, 1940, 1950, 1960), 
> right=FALSE,
>                     labels=c("1900-1929", "1930-1939", "1940-1949", 
> "1950-1959"))
> ## Compute the association with this new variable
> da <- dissassoc(biodist, biofam$cohort, R=1000)
> ## Printing results
> ## Differences are highly significant
> print(da)
>
>
> ## Second solution : Use the tree procedure
> ## It will automatically find the best binary splits
> biotree <- seqtree(biofam.seq~birthyr, data=biofam, diss=biodist)
>
> ##Printing the tree
> print(biotree)
> ## Displaying the tree (adjusting legend fontsize otherwise it's too big)
> ## You will need to install GraphViz for this
> seqtreedisplay(biotree, type="d", legend.fontsize=2)
>
>
> ## Creating a new cohort covariate according to the splitting points 
> found with the tree procedure
> biofam$cohort2 <- cut(biofam$birthyr, c(1900, 1929, 1941, 1947, 1951, 
> 1970), right=FALSE,
>                     labels=c("<=1928", "1929-1940", "1941-1946", 
> "1947-1950", "1951+"))
>
> ## Computing association with this new variable
> da2 <- dissassoc(biodist, biofam$cohort2, R=1000)
> ## Printing results
> ## Pseudo R2 is slightly higher than before
> print(da2)
>
> Hope this helps.
>
> Matthias Studer
>
>
>
> Le 30.10.2011 02:04, Zuluaga, Juan a écrit :
>
> Hello Traminer people,
>
> I read your Sociological Methods and Research paper.  The McVicar and 
> Anyadike-Danes (2002) dataset that you used has categorical covariates.
>
> How do you deal with quantitative variates?
>
> Thank you!
>
> -juan zuluaga
>
>
>
> _______________________________________________
> Traminer-users mailing list
> Traminer-users at lists.r-forge.r-project.org  <mailto:Traminer-users at lists.r-forge.r-project.org>
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/traminer-users
>
>
>
> _______________________________________________
> Traminer-users mailing list
> Traminer-users at lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/traminer-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/traminer-users/attachments/20111105/1b1de2e9/attachment.htm>


More information about the Traminer-users mailing list