Count Company's number of patent applications

[nico.rasters]

I found 2612 patents for Robert Bosch GmbH for 2006 (PATSTAT October 2014). The patent data was very clean. I could only find "Robert Bosch GmbH" with person_id 5235181 and doc_std_name_id 401. I searched in the full set of applications with authority office "DE" and filing year 2006. Restricted by appln_kind="A" later on.

There was no meaningful way for me to reduce the number to 2202 (=DPMA). However, the DPMA Annual Report 2006 was written in 2007. Too early, perhaps? Their list contains patent applications published by the German Patent and Trade Mark Office in 2006. The PATDPA patent database served as the data source.

Maybe someone who has access to PATDPA can run the query again and see if 2007 was indeed too early. If not, then PATDPA has a different coverage than PATSTAT... which raises a new set of questions.

With regards to accessing PATDPA I came as far as http://rzblx10.uni-regensburg.de/dbinfo ... el_id=1529

[Geert Boedt]

Hello Julia et al,
A couple of remark from my side.
Doing pre & post analysis on patent portfolios is very difficult only using (PATSTAT) bibliographical patent data for the following reasons:
a) patent data is primarily based on patent publications, meaning: the "post publication" status (also after grant) has to be analysed using (national) patent registers. In PATSTAT terms: analysing PRS codes in the tls221_inpadoc_prs for the respective countries. (not easy for non-experts)
b) because (in most countries) there is no obligation for the acquirer to register a change of ownership. Result: patent registers do not necessarily show the real situation.
c) as pointed out: "name harmonisation" is an issue. The eee-ppat (hrm_l2 name) effort is very useful but not perfect. In your case, if you only look at DE applications, it will probably not be much of a problem because DE data quality is good and "uniform" from the onset.

About comparing/replicating national statistics with PATSTAT. It is nearly impossible to get exactly the same numbers, unless you know exactly how the figures are obtained. Possible variations are due to: including PCT applications or not, including withdrawn applications or not (before publication!), including national phases of granted EP's (and PCT's) or not,... The best approach is to define your own methodology and your own conditions and take it from there. My approach when I see un-explainable large differences is to look at a small sample of detailed data, and identify what is missing. Once you have examples, then you can check why they are (ex)-included in your data sample.
Example: I used following query in depatisnet:

Code: Select all

((PA=(robert (L) bosch)))  AND ((AD > 31.12.2006) and (AD < 01.06.2007)) and PC =(DE) and pcod= (A1)

--> result= 808 hits.
In PATSTAT:

Code: Select all

SELECT distinct tls211_pat_publn.publn_auth+tls211_pat_publn.publn_nr+tls211_pat_publn.publn_kind publication ,tls201_appln.appln_auth+tls201_appln.appln_nr+tls201_appln.appln_kind application, tls201_appln.appln_filing_date,tls201_appln.appln_nr_epodoc, person_name
FROM tls206_person join tls207_pers_appln on tls206_person.person_id = tls207_pers_appln.person_id
  join tls201_appln on tls207_pers_appln.appln_id = tls201_appln.appln_id
  join tls211_pat_publn on tls201_appln.appln_id = tls211_pat_publn.appln_id
  where person_name like '%robert bosch gmbh%'
  and applt_seq_nr > 0
  and tls201_appln.appln_filing_date between '2007.01.01' and '2007.05.31'
  and tls201_appln.appln_auth = 'DE' and tls201_appln.appln_kind = 'A'
  and publn_kind = 'a1'
  order by tls201_appln.appln_filing_date asc

Result: 806 hits.
I tried a couple of other queries in depatisnet to get the official figure from the annual report (2202 for Bosch in 2006), but could not get much closer then what you had.
Only the persons who created the DE annual report figures could tell you the methodology and whether it is possible to replicate the figures.
About the double counting, as long as you count "distinct(appln_id)" you will never have duplicates at application level.

[Tim Grünebaum]

Hello Geert and Nico,

thanks for those useful replies!

I think it all comes down to the fact mentioned by Geert. It is nearly impossible to replicate the patent counts of a national patent office without knowing the exact procedure, code and database they used. However, this might not be as bad as I thought it was when I first came up with this issue.
With the codes we have written so far we should be able to get the relevant information for a specific research question. I am interested in a CDM-model approach (Crépon, Duguet, Mairesse 1998) by linking firms innovation expenditures to patent applications (maybe grants) and later profits. Maybe restricting the query to applications made in 'DE' is not a good idea as I have worldwide profits.

The DPMA says they are not using DEPATISnet (which in return also uses DOCDB) as their source for the annual reports. This is another reason why understanding their counts is cumbersome. But DEPATISnet gives a good tool for checking PATSTAT codes.

@Geert: Would you recommend including publn_kind = 'a1' and for what exact reason?
concerning a) Aren't all patents listed in PATSTAT are after publication?
(referring to the second post in post15465.html#p15465)
And can't we just check post-publication status like granted patents by including TLS201_APPLN.GRANTED = 1 ?
I also think name harmonisation should be fine in general. But I will check some other companies in PATSTAT and DEPATISnet as you did.

@Nico: There are indeed changes over time in the application counts. So if we execute the same query about 2006 in 2007 and in 2013 we will get different numbers. But the DPMA says that those changes are very small and could not explain the differences between their annual report and PATSTAT. So it seems to be a completely different counting method.
E.g. I suspect them to exclude PCT applications in general. Are we doing the same if we restrict our query to TLS201_APPLN.APPLN_KIND LIKE 'A' ? Maybe we should also include 'W'.
I do nt have access to PATDPA either, nor do I know anyone who has. I think this might go to far.

And you both are right. I we want to know the specific number of applications done by a specific company, we have to go into the company's history and nombers in detail. As I am interested in counting thousands of firms (knock on wood) this is beyond my ressources.
I think one can do well by using a reasonable code with a suitable name harmonisation and should not bother other pblished numbers because methods can be very different. I'll go ahead in doing so now