ApolloChoiceModelling forum

I have two small questions about deltaMethod_settings:

1) Is there any way to get it to report p-values? If not, how do I use the bootstrap method to calculate confidence intervals? (I've read this is a possibility on the forum but can't figure out how it works based on the manual.)
2) If I want to test whether two coefficients in a latent class analysis are statistically different, can I include something like (isdiff = "beta_systemcost_a+beta_systemcost_c") in the list for the deltaMethod_settings? Or is there another way I should be doing it?

Thanks,

Mara

Mara

1. the function currently reports standard errors and t-ratios. From the t-ratio, you can of course easily calculate the p-value. I've added the feature now to do this automatically from version 0.2.8 onwards, but for now, you can easily do it manually too. Just use (1-stats::pnorm(abs(trat))) where trat is the value you get for the t-ratio. This is for a one-sided p-value, so you can multiply the outcome by 2 if you want two-sided

For confidence intervals, you can use the standard errors given by the delta method, so you can use value+-1.96*se, for example

2. To know whether two parameters are different, you just need to test if their difference is different from 0. So e.g. expression=c(difference="beta_systemcost_a+beta_systemcost_c")

Stephane

Thanks Stephane.

For the second question, do you mean I put that in the list for the deltaMethod_settings? Hence using a Delta method approach? Or is there a way to conduct a joint F-test using a command like linearHypothesis?

Yes, using a Delta method. This is the standard approach, and is an exact calculation reflecting the maximum likelihood estimate properties - see http://www.stephanehess.me.uk/papers/jo ... B_2012.pdf

Yes, that's what I was thinking, thanks for clarifying.