NBER Patent Data Project in Stata

[MCGNL]

Hello,

I am trying to match compustat data with patent data by following this guideline: http://users.nber.org/~jbessen/matchdoc.pdf
Anyone did this before?

Currently I have a file with financial data sorted by gvkey and year.
Furthermore, I have downloaded the pdpcohdr and dynass files from the site: https://sites.google.com/site/patent...Home/downloads
And I downloaded the patent data file from http://elsa.berkeley.edu/pub/users/bhhall/NBER06.html.

The matchdoc.pdf file contains STATA code to merge the two data sets.
Two questions:

1) Looking at the patent data file above, I have multiple pdpasses and IPCs for each patent. For example:
year patent icl pdpass
1974 3930732 G01B 1500 10030734
1974 3930732 G01B 900 10030734

I want to create the variable npat, which containts number of patents for each pdpass-year. In the example, it should count as 1 patent for pdpass-year 10030734-1974. How do I create this in stata?

2) The stata example code says:

* now find the appropriate gvkey to assign the patents
gen gvkey=.
forvalue i=1/5 {
replace gvkey = gvkey`i' if gvkey`i'~=. & year>=begyr`i' &
year<=endyr`i'
}

When I perform this command I get the error code
invalid syntax
r(198);

I have the variables gvkey1 to gvkey5, begyr1 to begyr5, endyr1 to endyr5 (all from dynass file) and renamed appyear to year.
What could be wrong here?

Thanks in advance!

Best,
MCG

[Nick Cox]

Cross-posted at http://www.talkstats.com/showthread....oject-in-Stata

See the Advice in the FAQ about our policy on cross-posting.

[MCGNL]

Thank you, Nick.

Let me explain my first question more in general.
So let's say I have the following data (+ some more variables not relevant here):

Year Patent Pdpass
1977 4166686 6361394
1977 4171891 6361394
1977 4171891 6361394
1977 4166678 6361394
1977 4175847 6361394
1977 4166678 6361394
1977 4166682 6361394
1977 4146320 6361394

Where year = application year, patent = patent number and pdpass = a unique company identifier which later can be linked to its appropriate gvkey to match it with the compustat data.
As you can see, there are some doubles since they could have different international classification numbers. But I do want to count them as 1.
What I would like to have now is a variable called npat (number of patents) for each pdpass-year. Such that I get:

Pdpass Year npat
6361394 1977 6
1234567 1977 4 (for example)
....
....

How could I create this?
Thanks!

Best,
Benno

[Nick Cox]

help contract

[MCGNL]

But if I use the contract option it deletes every other variable and it gives the frequency of each unique patent number (instead of giving the number of patents per pdpassyear).
What can I do to keep the variable 'appyear', while getting one row per pdpassyear with the number of patents?

[Nick Cox]

contract is a command, not an option. The very first example in the help (did you try it?) shows how to count cross-combinations of two variables.

You don't give your code, but you presumably have in mind

contract pdpass

What you need is evidently

contract pdpass appyear

[Roberto Ferrer]

MCGNL,

(I just answered to this in the cross-posting given by Nick Cox.)

I believe you want to count distinct values, so you can drop the duplicates using duplicates drop before contracting.

More in http://www.stata.com/support/faqs/da...-observations/

Please honor the petition in the FAQ to use your real full name.