Errror message: Cholmod warning 'not positive definite' and NAs produced

[courtneygoodridge]

I have the following model:

mod.1 <- glmer(lane_position ~ heading + (heading | pNum), family = Gamma(link = "identity"), nAGQ = 0, data = magnitudedata)

The observed effect is:

fixef(mod.1)["heading"] heading 0.0491085

And then I run the following power analysis:

fixef(mod.1)["heading"] <- 0.01

powerSim(mod.1, nsim = 50)

However the result states I have zero power and 50 errors:

Power for predictor 'heading', (95% confidence interval):=============================================================================================================================| 0.00% ( 0.00, 7.11)

Test: z-test Effect size for heading is 0.010

Based on 50 simulations, (46 warnings, 50 errors) alpha = 0.05, nrow = 4052

Time elapsed: 0 h 0 m 10 s

Looking at the warnings and I see a message like this:

lastResult()$warnings

    stage index                                                                                  message

1 Simulating 1 NAs produced 2 Simulating 2 NAs produced 3 Simulating 3 NAs produced 4 Fitting 3 Cholmod warning 'not positive definite' at file:../Cholesky/t_cholmod_rowfac.c, line 431 5 Fitting 3 Cholmod warning 'not positive definite' at file:../Cholesky/t_cholmod_rowfac.c, line 431

Do you have any ideas on what this means and how to fix it? Help is much appreciated it, thank you!

[pitakakariki]

I'm guessing this is because you're using a Gamma distribution with an identity link. If your linear predictor becomes negative, then you end up in an invalid region of the parameter space.

[courtneygoodridge]

That does make sense. Thank you for the swift reply! Using the log link removes issue.