[R] fitting structured conditional (subset) models with loglm

[Michael Friendly]

With data like the following, a frequency table in data frame form, I'd 
like to fit a collection of loglm models
of independence of ~ attitude + memory for each combination of education 
and age.
I can use apply() if I first convert the data to a 2 x 2 x 3 x 3 array, 
but I can't figure out  an
equivalently simple use of an apply() approach with the data frame form.

 > library(MASS)
 > data("Punishment", package = "vcd")
 > str(Punishment)
'data.frame':   36 obs. of  5 variables:
  $ Freq     : num  1 3 20 2 8 4 2 6 1 26 ...
  $ attitude : Factor w/ 2 levels "no","moderate": 1 1 1 1 1 1 1 1 1 1 ...
  $ memory   : Factor w/ 2 levels "yes","no": 1 1 1 1 1 1 1 1 1 2 ...
  $ education: Factor w/ 3 levels "elementary","secondary",..: 1 1 1 2 2 
2 3 3 3 1 ...
  $ age      : Factor w/ 3 levels "15-24","25-39",..: 1 2 3 1 2 3 1 2 3 
1 ...

 > pun <- xtabs(Freq ~ memory + attitude + age + education, data = 
Punishment)
 >
 > mods.list <- apply(pun, c("age", "education"), function(x) 
loglm(~memory + attitude, data=x))
 > GSQ <- matrix( sapply(mods.list, function(x)x$lrt), 3, 3)
 > dimnames(GSQ) <- dimnames(mods.list)
 > GSQ
        education
age     elementary  secondary       high
   15-24   4.639061 0.08066111 0.09354563
   25-39  10.441996 0.96287690 0.48273162
   40-    12.680802 6.71016542 3.58752829
 > sum(GSQ)
[1] 39.67937

With the data in data frame format, I can do the same using the subset= 
argument, and
a series of separate calls (or for loops), but I'd rather us an apply()  
(or plyr) approach.

 > mod.1 <- loglm(Freq ~ memory + attitude, subset=age=="15-24" & 
education=="elementary", data=Punishment)
 > mod.2 <- loglm(Freq ~ memory + attitude, subset=age=="25-39" & 
education=="elementary", data=Punishment)
 > mod.3 <- loglm(Freq ~ memory + attitude, subset=age=="40-"   & 
education=="elementary", data=Punishment)
 > mod.4 <- loglm(Freq ~ memory + attitude, subset=age=="15-24" & 
education=="secondary", data=Punishment)
 > mod.5 <- loglm(Freq ~ memory + attitude, subset=age=="25-39" & 
education=="secondary", data=Punishment)
 > mod.6 <- loglm(Freq ~ memory + attitude, subset=age=="40-"   & 
education=="secondary", data=Punishment)
 > mod.7 <- loglm(Freq ~ memory + attitude, subset=age=="15-24" & 
education=="high", data=Punishment)
 > mod.8 <- loglm(Freq ~ memory + attitude, subset=age=="25-39" & 
education=="high", data=Punishment)
 > mod.9 <- loglm(Freq ~ memory + attitude, subset=age=="40-"   & 
education=="high", data=Punishment)
 >
 > mod.list <- list(mod.1, mod.2,mod.3, mod.4, mod.5, mod.6, mod.7, 
mod.8, mod.9)
 >
 > GSQ <- matrix( sapply(mod.list, function(x)x$lrt), 3, 3)
 > dimnames(GSQ) <- list(age = levels(Punishment$age),
+                       education = levels(Punishment$education)
+                       )
 > GSQ
        education
age     elementary  secondary       high
   15-24   4.639061 0.08066111 0.09354563
   25-39  10.441996 0.96287690 0.48273162
   40-    12.680802 6.71016542 3.58752829
 >

-- 
Michael Friendly     Email: friendly AT yorku DOT ca
Professor, Psychology Dept. & Chair, Quantitative Methods
York University      Voice: 416 736-2100 x66249 Fax: 416 736-5814
4700 Keele Street    Web:http://www.datavis.ca
Toronto, ONT  M3J 1P3 CANADA