deltaMethod() function not allowing inclusion of dlogis(). Any way to get around this?

by

I was originally running the following code and it was working great:

avg_x1 <- mean(df$x1)
avg_x2 <- mean(df$x2)
avg_x3 <- mean(df$x3)

het_prob <- hetprob(y ~ x1 + x2 + x3 | x1, link="probit", data=df)
test <- deltaMethod(het_prob, "dnorm( (Intercept) + avg_x1*x1 + avg_x2*x2 + avg_x3*x3)")

I now want to tweak this as follows:

avg_x1 <- mean(df$x1)
avg_x2 <- mean(df$x2)
avg_x3 <- mean(df$x3)

het_log <- hetprob(y ~ x1 + x2 + x3 | x1, link="logit", data=df)

test <- deltaMethod(het_log , "dlogis( (Intercept) + avg_x1*x1 + avg_x2*x2 + avg_x3*x3)")

However, I’m getting the error:

“Error in D(g., para.names[i]) :
Function ‘dlogis’ is not in the derivatives table”

Is there any workaround to replace dnorm with dlogis in deltaMethod()?

  • Can you provide a minimal reproducible example? Is this deltaMethod from the car package …?

    – 

  • Actually, looks like hetprob is from the Rchoice package … ??

    – 

  • From the error message, you would need to add dlogis to the “derivatives table”, but that’s built in to R, so this isn’t easy. The nlsr::nlsDeriv() function allows you to add your own derivatives, but then you’d need to modify deltaMethod() (or something it uses) to use nlsDeriv() instead of D().

    – 




  • There’s also a Deriv package that similarly allows the derivative table to be extended (but deltaMethod would still need to be rewritten)

    – 

  • Yes, deltaMethod is from the car package and hetprob is from the Rchoice package. I can look into the options you mention, thank you!

    – 

Leave a Comment