Calculating citations index - NBER Database

[Anton Astakhov]

Hello,

I am working with NBER Patents Database, and I am trying to construct so-called Citation Index (CI) following Hirschey, M., Richardson J.V. (2004) Are scientific indicators of patent quality useful to investors? CI is defined as "the number of citations generated in the current year by patents granted to the company during the most recent 3-year period", including current year. I have a balanced panel, where one panel in general looks like this:

id	patent_no	grant_year	year	citations
10300	4146300	1985	1985	0
10300	4146300	1985	1986	2
10300	4146300	1985	1987	2
10300	4146300	1985	1988	7
10300	4146300	1985	1989	5
10300	4146301	1986	1985	0
10300	4146301	1986	1986	0
10300	4146301	1986	1987	4
10300	4146301	1986	1988	1
10300	4146301	1986	1989	3

where id is a unique identificator of the firm, patent_no is a number of patent, grant_year is the year, in which Patent Office approved firm's application for the patent, and citations is the number of citations, received by the patent in year i.

So I need to generate some variable CI, which in this particular case would look like (if we assume that firm 10300 obtained only 2 patents in its lifetime):

id	year	CI
10300	1985	0
10300	1986	2
10300	1987	6
10300	1988	1
10300	1989	3

E.g., for year 1987 CI = 2 (because patent no 4146300 has been granted in 1985, so less or equal than 3 years ago) + 4 (because patent no 4146301 has also been granted within last 3 years) = 1.

I can't figure out how to do this, so any help would be much appreciated. Actual dataset is large, with year=1975..2006, and number of patent_no-year observations almost 100 mil.

Thank you!

[Oded Mcdossi]

I didn't manage to exactly reproduce your second table, especially for year 89 for patent No. 301, but you can modify the example to match your desire.

​
bys id patent_no (year): g ci1=cond(inrange(year-grant_year,0,2), citations,0)​
collapse (sum) CI= ci1, by( id year)
list

[Anton Astakhov]

From what I can tell this works perfectly without any modifications, thank you Oded! It doesn't reproduce for year 1989 simply because I made a mistake in the second table: by 1989 patent 301 has already been granted more than 3 years ago, so CI should be 0 actually; sorry for confusion.