Re: st: filling in the gaps

[Nick Cox <njcoxstata@gmail.com>]

Maarten and Jan gave various good advice, but there is a more
optimistic take. Use linear interpolation (strictly  extrapolation) on
-year- and -grade-. Presumably the usual pattern is to advance a grade
each year. If that is inconsistent with the rest of the evidence the
result will be grades with fractional parts.

The command is -ipolate-

Some kind of sensitivity analysis is probably still a good idea,
notably to compare model results for the dataset with interpolation
with those for the conservative dataset, with only definitely known
grades and years.

All that said

Nick

On Fri, Jan 18, 2013 at 9:14 AM, Maarten Buis <maartenlbuis@gmail.com> wrote:
> --- Am 18.01.2013 09:21, schrieb John Singhammer:
>>> I'm working on a dataset consisting of information on 45.000 school
>>> children Data has been collected annually since 2009, though not
>>> for all children Information on grade has been imported from a
>>>  national register. However, that information is only available up to
>>> the year 2011
>
> --- On Fri, Jan 18, 2013 at 9:44 AM, Jan Ditzen wrote:
>> if I understood your problem correctly the following should help:
>>
>> by id (sch_year), sort: replace grade = grade[_n-1]+1 if grade == .
>
> Technically that is true, but statistically that is typically bad
> practice. This way you impose a very severe pattern on the grade
> profiles of those kids. If that is what you want to study, than any
> subsequent analysis is no longer empirical research but just
> reproducing your assumptions.
>
> In general I would say that if you only have data on a key variable
> till 2011, than that is it: you have data till 2011 and no more. If
> you really really need those subsequent years and you really really
> cannot wait till those data become available than you could try
> multiple imputation (type in Stata -help mi-). However, given the fact
> that these are complete years that are missing I would strongly
> recommend against that. Instead I would just stick to the years
> 2009-2011 and in a couple of years, when the data for 2012 and 2013
> become available, write a new article for the period 2009-2013.
>
> -- Maarten
>
> ---------------------------------
> Maarten L. Buis
> WZB
> Reichpietschufer 50
> 10785 Berlin
> Germany
>
> http://www.maartenbuis.nl
> ---------------------------------
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/