Closed rbcavanaugh closed 3 years ago
there is another, related issue here - how can we show an estimate for the distributions for each ability estimate? The fullDist function in catR doesn't seem to want to work. can we run a quick model using rstanarm? or should we just show a normal distribution where the distribution is Normal(ability_estimate, 1.96*SEM)?
I think Will should chime in, but I think that the ~N(θ, 1.96*SEM|θ) should work just fine.
Yeah, if we're talking about the posterior distribution of an individual score estimate, I think the normal distribution is just fine. But if we're talking about plotting the current examinee's point estimate or posterior distribution in relation to the calibration sample, that's another matter, because as I commented in the other thread about scaling, the scores do not seem to be normally distributed in relatively large samples. What if you had an actual histogram of the score estimates from the calibration sample with the point estimate and 95%CI superimposed or printed beneath. You could display the CI as simple error bars, or, as i think you're suggesting, a normal density.
I agree with Will.
Doable - send me a .csv with a vector of all the calibrated sample estimates
waiting on t-score scaled theta estimates.
thetas_Tscaled_MAPPDn296_R03n39_2021_10_04.csv Here you go. This is for the 335-person sample. I hope to able to update this file and the item parameters with the additional PSU/UW cases that will bring it north of 350.
perfect thanks. I'll use this to generate a histogram behind the estimate on teh results page. thanks!
uploaded.
Results plot needs a few updates.