As you see, I have more than 500 groups of data. how can I get the std.coef in mult-linear regression for each group? Until now, I only know how to get one group by using lm.beta. For example:
library(QuantPsyc)
p <- read.csv('Data1.csv')
model1 <- lm(YCOORD110 ~ TEMMIN + TEMVAR)
lm.beta(model1)
For groups, several codes like by, apply can be used, but seem much more complicated beyond my control.
very appreciate for your answers!
Use a for loop:
library(QuantPsyc)
attach(iris)
group <- levels(Species)
group
out03 <- vector()
for (i in group){
out00 <- subset(iris, Species==i)
out01 <- lm(Petal.Width ~ Petal.Length + Sepal.Width, data=out00)
out02 <- lm.beta(out01)
out03 <- c(out03, out02)
}
finalTable <- cbind(group, data.frame(matrix(out03, ncol=2, byrow=T)))
colnames(finalTable) <- c("Species", "Petal.Length beta", "Sepal.Width beta")
finalTable
Final tabulated output:
Species Petal.Length beta Sepal.Width beta
1 setosa 0.2997348 0.1794891
2 versicolor 0.6043644 0.3252391
3 virginica 0.1268589 0.4868520
As far as I can understand the query is how to find group coefficients in a single regression model, where some independent variables may be dependent on group (M/F, let us say). In this case you need to create n dummy binary variables (for n groups) as separate independent variables. These variables will have the values 1 for all individual observations of their respective group, and 0 values for all other observations. Now these dummy variables will be multiplied with each independent variable and included in the linear regression model.
E.g. Lets say your model is: (Y = C + A.x1 + B.x2) and we have 2 groups.
We will create 2 dummy variable, z1 and z2. z1/z2 is a binary variable with value =1 for all observations in the respective group and rest values will be 0. Now input your linear regression model as: Y = C + A1.x1.z1 + A2.x1.z2 + B1.x2.z1+ B2.x2.z2
A1, A2 will respectively, the coefficients for groups 1 and 2
