How to impute a variable from plausible values in PISA data

[Minda Tan]

It's my first time using PISA data. I have already read the help file of "pv". My question is can I use it to generate an imputed variable representing the plausible values?

pv, pv(pv*math) : gen math=@pv

I tried with the code above, and stata gave me one. However, this variable is equal to pv1math. Is there another package that can help me with this issue? If not, how should I use the chained multiple imputations to achieve my goal?

Thank you!!

[ericmelse]

I am not certain if this might be of use, but have a look at the user community contributed module uirt, created by Bartosz Kondratek, which facilitated plausible values:

ssc install uirt , replace

A paper about the module was recently published by Bartosz:
Kondratek, B. (2022). uirt: A command for unidimensional IRT modeling. The Stata Journal, 22(2), 243–268. https://doi.org/10.1177/1536867X221106368

[Minda Tan]

I am not certain if this might be of use, but have a look at the user community contributed module uirt, created by Bartosz Kondratek, which facilitated plausible values:

ssc install uirt , replace

A paper about the module was recently published by Bartosz:
Kondratek, B. (2022). uirt: A command for unidimensional IRT modeling. The Stata Journal, 22(2), 243–268. https://doi.org/10.1177/1536867X221106368

Thank you for the reply! I tried it with pv*math. The State reported error:

J(): 3900 unable to allocate real <tmp>[606627,314632]
return_group_item_info(): - function returned error
uirt(): - function returned error
<istmt>: - function returned error

I will read the paper you recommended to find a solution. Thanks again!

[Philip Matthews]

From your posts it seems that you might misunderstand what values such as pv1math, pv2math etc. represent in the PISA databases. You can’t impute/produce a ‘better’ single set of scores that will replace the sets of five or ten plausible values. (That is what your piece of Stata code appears to be trying to do – apologies if I am wrong.) What one has to do is use the sets of pv values to estimate, for example, mean scores, regression coefficients, standard errors etc.

In case you haven't discovered it, the best starting place to learn about working with plausible values is the PISA Data Analysis Manual. There are two versions: one with SPSS, and one with SAS code. Don’t be put off by the fact that there is no Stata version — you don't need to know SPSS or SAS syntax. The underlying theory is the same in both and forms the basis of Stata user-written packages such as pv. Especially, see Chapters 4, 6, 7 and 8.

To get the manual, go to

https://www.oecd.org/pisa/

At bottom right of page click on ‘Publications’. Then search for PISA Data Analysis Manual and you should be able to download a pdf of the publication.

You may find two other user-written packages useful. In Stata, type

findit pisatools

findit repest

(repest is written by two OECD researchers.)

[Minda Tan]

From your posts it seems that you might misunderstand what values such as pv1math, pv2math etc. represent in the PISA databases. You can’t impute/produce a ‘better’ single set of scores that will replace the sets of five or ten plausible values. (That is what your piece of Stata code appears to be trying to do – apologies if I am wrong.) What one has to do is use the sets of pv values to estimate, for example, mean scores, regression coefficients, standard errors etc.

In case you haven't discovered it, the best starting place to learn about working with plausible values is the PISA Data Analysis Manual. There are two versions: one with SPSS, and one with SAS code. Don’t be put off by the fact that there is no Stata version — you don't need to know SPSS or SAS syntax. The underlying theory is the same in both and forms the basis of Stata user-written packages such as pv. Especially, see Chapters 4, 6, 7 and 8.

To get the manual, go to

https://www.oecd.org/pisa/

At bottom right of page click on ‘Publications’. Then search for PISA Data Analysis Manual and you should be able to download a pdf of the publication.

You may find two other user-written packages useful. In Stata, type

findit pisatools

findit repest

(repest is written by two OECD researchers.)

Sorry to reply so late! Thank you so much!!