I have a data that looks like this
id var1 var2 var3 response var4
1 1 0 cat1 E 1 T4
2 1 0 cat1 E 1 T2
3 2 0 cat2 B 1 <NA>
4 3 0 cat3 C 0 <NA>
5 4 0 cat4 D 0 T3
6 4 0 cat4 D 0 T1
7 5 1 cat1 A 1 T4
8 5 1 cat1 A 1 T3
9 6 1 cat3 C 1 T3
10 7 0 cat1 C 0 T1
I want to run a glm for the dependent variable "response"
as you can see the ID's are not unique, that's because each ID can take more than one value for "var4" ,
to that end I tried model.matrix but it doesn't preserve the Id's, making it difficult to check the results,
I am sure there is a simple way to do this, can someone help?
to generate the above data you can use
# GENERATING THE FIRST DATA FRAME
set.seed(1984)
df1 <- data.frame(id = 1:15,
var1 = sample(0:1, 15, replace = T, prob = c(.7,.3)),
var2 = sample(c('cat1','cat2', 'cat3','cat4'),15, replace = T),
var3 = sample(LETTERS[1:5],15, replace = T ),
response = sample(0:1, 15, replace = T)
)
# GENERATING THE 2ND DATA FRAME
set.seed(1984)
df2 <- data.frame(id = sample(1:15, 20, replace = T),
var4 = sample(c('T1','T2','T3','T4'), 20, replace = T))
df2 <- unique(df2[order(df2$id), ])
row.names(df2) <- NULL
# MERGING THE TWO
df3 <- merge(df1, df2, by = 'id', all = T )
df3
I would reshape the data using dcast from the reshape2 package. This will reshape your data so each id is on one row and one-hot-encode var4 prior to feeding the dataframe into model.matix. Using your minimum working example this would be something like:
library(reashpe2)
newDF <- dcast(df3, ...~var4,function(x) length(x))
model.matrix(response ~., newDF)