plausible values of pisa

[dindays]

dear anyone, I would like to know if anyone can share with me scripts for running plausible values of PISA data.

1. I need to compute a single composite score from those five plausible values of each subject in PISA data and as I know from the oecd manual, we cannot just simply take the average of these fives. I applied to use the pv - commands. But I just need to make sure what is the precise way to get the single score of for example math for a country.

2. How to do a simple regression using these values?

Thank you so much in advance for the helps.

Regards,
Daim

[Maarten Buis]

The whole point of the plausible values is there there is not a single value, that is what the pv command (user written, available from SSC) is for. The first example in the help-file of pv answers your second question.

[Stephen Jenkins]

Daim: you're following in the footsteps of many previous PISA data users, so surely there is guidance 'out there' for you to follow regarding these basic questions -- over and above what is in the PISA manuals? Regardless, Maarten is absolutely right.
PISA plausible values are derived as predicted responses from an Item Response model fitted to the data and since the model is stochastic, there can be multiple predicted values. One way of thinking about this (and others more expert than I may correct me) is in terms of a missing data model and Multiple Imputation -- we do not observe the true score for a 15 year-old, rather a set of observed test scores. (The IR model relates the true (but latent) score to observed scores and observable and unobservable characteristics.) Given values of the model parameters, we can impute a score to individuals. Given model uncertainty, we could and should do this multiple times. It is analogous to Multiple Imputation. Indeed if you look at the code for pv.ado using viewsource pv.ado, you'll see some MI type computations being set up (averaging to get point estimates, and SEs derived using a combination-of-variances formula that is appears similar to the famous Rubin's Rules in MI.) As Maarten says, pvis a specialist program set up to handle this sort of situation; it seems tailor-made for what you want to do.

[dindays]

Hello Maarten and Stephen,

Thanks a lot for your suggestions. Will come back if I have questions more on this.

Cheers,
Daim

[Luis Muelle]

Hello friends,
I am working with PISA 2012 data and I would like to obtain high/ low achievers subpopulations. Each student has 5 plausible values scores in maths for
instance and it has been established that a 420 points score is the boundary separating a student from a high to a low score.
With this subpopulations (more than 420 score points and less than 420 score) I obtain a high/low dummy variable and I easily apply a melogit command
with this binary variable as dépendent variable. My question is How divide this scores of students in two subpopulations?.

[Stephen Jenkins]

Luis: your question is quite different from the earlier questions about plausible values. (The fact that you are using PISA data is the main connection, but insufficient.) Please re-post your question as a new topic (with an informative title) so that there is a separate thread.

[daoquendo]

Hello Daim, I would like to know if you find out the way to calculate a single composite score from those five plausible values of each subject in PISA. I need to do that for each student. I would like to obtain the corresponding PISA score on mathematics and reading for each student. Thanks very much

,DO

[daniel klein]

I need to do that for each student

Why?

Others have already pointed out that such approach is likely to be statistically unsound and inconsistent with the hole idea of plausible values.

Best
Daniel