search

craddm / metaSDT

A simple R package for calculating (meta-) SDT measures
Other
11 stars 2 forks source link

fit_meta_d_SSE producing very large meta_da values #5

Open kitdouble opened 4 months ago

kitdouble commented 4 months ago

Hi there,

When comparing results from fit_meta_d_SSE and fit_meta_d_MLE I sometimes get very different results, with very large meta_da values from fit_meta_d_SSE. For example, the below example will produce a meta_da value of 0.932558 from fit_meta_d_MLE but a value of 5 from fit_meta_d_SSE

nR_S1 <- c(0,  2, 16,  7,  0,  0,  0,  0, 10, 15,  4,  0)
nR_S2 <- c(0,  0,  0,  3,  1,  0,  0,  0, 10, 22, 18,  0)

metaSDT::fit_meta_d_SSE(nr_s1 = nR_S1, nr_s2 = nR_S2)
metaSDT::fit_meta_d_MLE(nr_s1 = nR_S1, nr_s2 = nR_S2)
craddm commented 1 month ago

Hi, that happens sometimes. IIRC MLE is a bit more robust than SSE. If you're getting a value of 5, that really means the optimisation process never really converged on a "good" value. It searches for the "best" value within a range of -5 to 5. The different methods sometimes throw up different results as they use different ways of determining what the best value is.