szorowi1 / mace-irt Goto Github PK

analysis

• make standalone scripts for fit indices & run
• verify scripts on simulated data
• CV-LOO transfer
• collect new mace data
• sit and think about how to handle missing siblings data (missing sibling data on 16% of participants; where available, only 10% of participants have no siblings; suggest that only 10% of these participants (1.6% overall) may be biased structural zeros. in this case, simply include all responses from participants with missing data.)
• confirm scoring for reverse-coded items (we're scoring correctly.)
• dif for items with rare endorsement? (maybe there's a robust literature, but not immediately finding something. that said, simulations reveal the obvious: with small samples comes noisiness, so care should be taken in items with low overall endorsement)
• collapse item #6
• re-fit models
• BRMSEA? (https://doi.dox.org/10.1177/0013164417709314, https://doi.org/10.1080/00273171.2014.911075)

writing

• methods: model descriptions
• methods: model fitting
• methods: discrepancy measures
• table #2: update item numbers