effect_robustness: Robustness of treatment effects

In knickodem/WBdif: Investigate DIF and generate reports

Robustness of treatment effects

Description

Compares standardized treatment effects estimated with and without adjustments for DIF.

Usage

effect_robustness(dif.models, irt.scoring = "WLE")

Arguments

dif.models	Output from dif_models
irt.scoring	What type of IRT scoring procedure should be used? Passed to the method argument of fscores. See help(fscores, mirt).

Details

Data and models are passed to effect_robustness via the dif.models argument. If tx.group.id == dif.group.id, then the unconditional standardized treatment effect is computed. If tx.group.id != dif.group.id, the standardized treatment effect is computed conditional on dif.group.id (e.g., conditional on gender), and the difference in treatment effects is also reported. The treatment effects and their standard errors are computed using the method described by Hedges (2007).

Treatment effects and their standard errors are reported for four different outcome variables.

• The unit-weighted total score computed with all items.

• The unit-weighted total score computed with DIF items omitted (i.e. the items identified in dif.models$biased.items).

• IRT scores computed using a model that constrains all items to have equal parameters over levels of dif.group.id (i.e., dif.models$no.dif.mod) .

• IRT scores computed using a model that allows parameters of DIF items to vary over levels of dif.group.id (i.e., dif.models$dif.mod).

IRT scores are computed using the method given by irt.scoring.

Value

A list of data.frames, with each data.frame summarizing the standardized effect size and standard error for the four different outcomes. The length of the list depends on whether the conditional or unconditional effect sizes were requested.

References

Hedges, L. V. (2007). Effect Sizes in Cluster-Randomized Designs. Journal of Educational and Behavioral Statistics, 32, 341–370. https://doi.org/10.3102/1076998606298043.

Examples

data("mdat")

# prep data
dif.data <- dif_data_prep(item.data = mdat`[`5:ncol(mdat)],
                             dif.group.id = mdat$gender,
                             na.to.0 = TRUE)

# With user-specified biased items
dif.models <- dif_models(dif.data = dif.data, biased.items = c(1, 5, 7))
effect_robustness(dif.models = dif.models, irt.scoring = "WLE")

knickodem/WBdif documentation built on Feb. 3, 2024, 2:20 a.m.