Data frame of mixed logit log-likelihoods as the number of draws grows

This data frame contains the converged log-likelihoods of the same preference space mixed logit model used in the runtimes and runtimes_draws benchmarks, estimated repeatedly at increasing numbers of random draws. At each draw count the model is estimated 10 times using randomized MLHS draws with a different seed each time, with every run starting from the same converged reference solution, so the across-seed spread in log-likelihood at a given draw count reflects simulation error: the noise in the simulated log-likelihood surface shifts the converged value, and at low draw counts it can distort the surface enough that runs converge to different local optima entirely. It is used to show that low draw counts produce unstable simulated log-likelihoods, which stabilize as the number of draws grows. It is generated by the script data-raw/runtimes.R in the package repository.

Usage

data(loglik_draws)

Format

Variable	Description
numDraws	The number of random draws used during estimation.
seed	The random seed used to generate the MLHS draws.
logLik	The converged log-likelihood of the model.
time_sec	The estimation time in seconds (on a 10-core Apple M-series Mac).

Source

The data-raw/runtimes.R script in the logitr repository.

Examples

data(loglik_draws)

head(loglik_draws)
#> # A tibble: 6 × 4
#>   numDraws  seed logLik time_sec
#>      <dbl> <int>  <dbl>    <dbl>
#> 1       50     1 -1271.    0.274
#> 2       50     2 -1249.    0.281
#> 3       50     3 -1273.    0.280
#> 4       50     4 -1248.    0.274
#> 5       50     5 -1274.    0.279
#> 6       50     6 -1273.    0.278