Insight into different types of patent families

Catalina Martinez, from IIP - Madrid recently published in OECD's STI working paper series a document titled insight into different types of patent families" containing an exaustive overview of the different patent families definitions,  an impact analisys of adopting one definition or another and a deep numeric analisys based on patstat.
One citation: "most definitions will give the same results for 75% of the families. In particular, equivalents, single-priority families and extended patent families will effectively provide the same results for families characterised by one single parent and direct subsequent filings."

Investigating patstat through application numbers

When investigating patstat by application number / application office, we soon come to a problem 'cause this pair is not a unique identifier.

FI if we would look into espacenet for application number 143781 we would get the following 7 results:

1    Molding for a front glass     Publication info: US5553428 (A) - 1996-09-10 2    Display tab for overwrapped package     Publication info: US4779733 (A) - 1988-10-25 3    Welding socket for thermoplastic materials, method and apparatus for its manufacture.     Publication info: DK143781 (A) - 1981-10-01 4    WIRBELBETTVERFAHREN ZUR DURCHFUEHRUNG VON ENDOTHERMEN CHEMISCHEN UND/ODER PHYSIKALISCHEN PROZESSEN     Publication info: BE814393 (A1) - 1974-08-16 5    Information handling apparatus     Publication info: US3248560 (A) - 1966-04-26 6    Magnetic unit     Publication info: US2163161 (A) - 1939-06-20 7    System of gas control     Publication info: US1348379 (A) - 1920-08-03

Still nothing bad, a part from the fact that when investigating the espacenet application numbers, in many cases the application year (and occasionaly a 0) is added; so if you look for patstat BE 143781 (that is BE814393 in above example) you'll find in espacenet web page

Application number: BE19740143781 19740430  

But having a deeper look into the results we will discover that we have 5 hits for USPTO, that means (using the same mechanism)

US19930143781 19931101
US19880143781 19880114 
US19610143781 19611009
US19370143781 19370520 
US19170143781 19170122

so at what extent duplication of application is spread?
by running a simple query like the one below we may get some results...

Select
  patstat.tls201_appln.APPLN_AUTH, 
  patstat.tls201_appln.APPLN_NR,
  Count(patstat.tls201_appln.APPLN_ID) As appcount
From
  patstat.tls201_appln
Group By
  patstat.tls201_appln.APPLN_AUTH, patstat.tls201_appln.APPLN_NR
Having
  Count(patstat.tls201_appln.APPLN_ID) > 1

What we get is that out of a table counting 63.595.305 applications, 5.805.769 are duplicated.
How... lets aggragate a bit the results....

count of duplications	count of appl #
2	5721284
3	83472
4	979
5	29
6	1
7	3
9	1
Total count:	5805769

(the lucky application with 9 occurrence is TW85219019)

Obviously by multipling the count of duplications for the number of occurrences we get
 11.697.081 applications involved in the issue...

Splitting by country and comparing it to the total number of applications by country we get that the distribution varies a lot from country to country: from the 0,03 of CN or 0,06 of EP up to the 50.73 % of SE

appln auth	# of duplicated ap nr	total # of applications	%
AM	2	147	1,36%
AP	12	4742	0,25%
AR	408	79581	0,51%
AT	6334	1028408	0,62%
AU	683801	1553990	44,00%
BA	44	344	12,79%
BE	4639	642989	0,72%
BG	11611	53602	21,66%
BR	58551	496326	11,80%
BY	20	761	2,63%
CA	2390	2580140	0,09%
CH	8672	1055204	0,82%
CL	10	4453	0,22%
CN	826	2803867	0,03%
CS	1765	166312	1,06%
CU	26	2843	0,91%
CY	2	2621	0,08%
CZ	5617	66921	8,39%
DD	8937	266209	3,36%
DE	1506427	6792342	22,18%
DK	12118	429570	2,82%
EA	32	13114	0,24%
EC	2	4949	0,04%
EE	1298	6251	20,76%
EG	160	11450	1,40%
EM	4	4254	0,09%
EP	1441	2388584	0,06%
ES	195371	902689	21,64%
FI	20051	271371	7,39%
FR	104899	2917043	3,60%
GB	133141	3319126	4,01%
GR	194	97444	0,20%
GT	4	1301	0,31%
HK	34	69450	0,05%
HR	154	11297	1,36%
HU	17796	137061	12,98%
IB	8	65369	0,01%
ID	4561	14768	30,88%
IE	773	91248	0,85%
IL	1155	160970	0,72%
IN	358	66238	0,54%
IS	57	7797	0,73%
IT	161988	708724	22,86%
JP	7567660	16282860	46,48%
KE	28	1392	2,01%
KR	279459	1659443	16,84%
KZ	4	477	0,84%
LT	22	3651	0,60%
LU	34	68491	0,05%
LV	26	4835	0,54%
MA	4	10092	0,04%
MC	682	2789	24,45%
MD	554	4586	12,08%
MK	2	87	2,30%
MN	12	246	4,88%
MX	1262	162007	0,78%
MY	26	11106	0,23%
NL	17313	607873	2,85%
NO	1610	226939	0,71%
NZ	30	109638	0,03%
OA	100	12934	0,77%
PE	8	431	1,86%
PH	339	23261	1,46%
PL	2263	233730	0,97%
PT	112	81034	0,14%
RO	472	61095	0,77%
RU	2117	394084	0,54%
SE	421795	831377	50,73%
SG	595	51063	1,17%
SI	235	17619	1,33%
SK	614	23261	2,64%
SU	3977	1249050	0,32%
TR	837	42800	1,96%
TT	2	52	3,85%
TW	2956	369739	0,80%
UA	748	50213	1,49%
US	434521	11376401	3,82%
UY	66	6573	1,00%
VN	8	240	3,33%
WO	39	5065	0,77%
XH	2	1242	0,16%
YU	91	33687	0,27%
ZA	749	256542	0,29%
ZM	4	2742	0,15%
ZW	10	2909	0,34%

 From table have been excluded 9779 applications from 86 lesser patent offices that have no duplications.


Thanks also to Elena Verdolini for rising and helping investigating this issue.

Counting patstat patents by application year

Here below a small table created by counting by application year patents applied @ EPO vs patents applied @ USPTO. Data have been counted from september 2009 data, starting from 1980.
Note that the peak is in 2004 for US and 2006 for EPO due to the examination lag (for both office) and also to the delay in including Us patents in patstat, for USPTO.
Nota also that 2001 sudden rise of 100% of patent count for USPTO is due to a change in the US Patent Law, since US Patent and Trademark Office was obliged to publish patent applications from 29 November 2000. However USPTO internal publishing system meant that publication actually began in March 2001.

appl year	EPO pat#	USPTO pat#
1980	19724	67106
1981	24959	64458
1982	28522	65614
1983	31609	62165
1984	36952	67690
1985	39375	72161
1986	43083	75925
1987	45817	82387
1988	52186	91028
1989	57727	97146
1990	63907	100332
1991	59280	101194
1992	60626	104798
1993	60066	109216
1994	62005	124374
1995	65308	145762
1996	71413	146351
1997	80488	172249
1998	90974	175349
1999	98718	193052
2000	109395	229731
2001	115110	428829
2002	112839	443539
2003	115652	458809
2004	120813	490649
2005	126739	488568
2006	129292	404225
2007	112413	328202
2008	35728	183275
2009	2909	13271

Here below a graph showing the differences...

Chinese patents data coverage in PATSTAT

Following my recent post about chinese patents data, I had a a deeper look into SIPO (chinese patent office) and organized data by year (see table below) and compared them against PATSTAT 10/2009 data.
We can understand that, leaving aside designs, that data coverage is usually over 90% on applied patents.

Year	Total	Inv + utility	Invention	Utility Model	Design	PATSTAT	% coverage
1985	14.372	13.732	8.558	5.174	640	12.532	91,26
1986	18.509	17.682	8.009	9.673	827	15.992	90,44
1987	26.077	24.765	8.059	16.706	1.312	22.411	90,49
1988	34.011	32.052	9.652	22.400	1.959	27.388	85,45
1989	32.905	30.386	9.659	20.727	2.519	27.215	89,56
1990	41.469	37.752	10.137	27.615	3.717	32.090	85,00
1991	50.040	44.705	11.423	33.282	5.335	37.713	84,36
1992	67.135	58.778	14.409	44.369	8.357	48.832	83,08
1993	77.276	67.117	19.618	47.499	10.159	54.916	81,82
1994	77.735	64.578	19.067	45.511	13.157	57.315	88,75
1995	83.045	65.377	21.636	43.741	17.668	60.224	92,12
1996	102.735	78.121	28.517	49.604	24.614	67.837	86,84
1997	114.208	83.795	33.666	50.129	30.413	73.285	87,46
1998	121.989	87.357	35.960	51.397	34.632	79.939	91,51
1999	134.239	94.186	36.694	57.492	40.053	91.110	96,73
2000	170.682	120.562	51.747	68.815	50.120	112.283	93,13
2001	203.573	142.926	63.204	79.722	60.647	132.884	92,97
2002	252.631	173.371	80.232	93.139	79.260	165.357	95,38
2003	308.487	214.433	105.318	109.115	94.054	204.918	95,56
2004	353.807	242.958	130.133	112.825	110.849	234.252	96,42
2005	476.264	312.893	173.327	139.566	163.371	286.538	91,58
2006	573.178	371.856	210.490	161.366	201.322	337.228	90,69
2007	693.917	426.485	245.161	181.324	267.432	354.732	83,18
2008	828.328	515.424	289.838	225.586	312.904	255.837	49,64

The delusion comes from the quality of data coverage...

When examining the applicants data out of 776607 applicant none has a zip code or a city.
Only 3161 have an address (where address city zip are packed altogether) but all of them are non-chinese.
So no geographic analisys is possible.

One more thing to be careful about are counts by publication numbers: FI if you look for patent # CN1237182  you will find 2 different patents: the first with status A (unexamined) and the latter with status C (granted).

So we should clarify the meaning of Publication number. According to SIPO, there is more than one kind of publication numbers which are reflected by the Kind Code:

  ‘A’: Publication of Unexamined application for patent for invention

  ‘C’: Publication of Granted patent for Invention

‘Y’: Publication of Granted patent for Utility Model

‘D’: Publication of Granted patent for Design

Last but not least: IPCs.

Out of 3.270.000 published patents 1.136.000 have no indication of primary ipc class.

So we're going to use data straight from SIPO to build a chinese patents database...

(bytheway, if you are willing to search into sipo website, this is the link to english search page:
http://218.240.13.210/sipo_EN/search/tabSearch.do?method=init )

IPC: core vs advanced level in patstat

As very vell explained from EPO website (here) IPC classification contains over 70.000 entries, that give too much detail for most users.
That's the reason why aside from this advanced classification has been introduced a core classification containing "only" 20.000 entries.
The advanced level is a superset of the core level, ie it consists of all core level groups and many finer subdivisions, allows updates to be made every three months, which means it will grow considerably in the years ahead.
In PCT minimum documentation have an advanced-level IPC. They also have a core level IPC that is derived from the advanced level.

In patstat we can distinguish advanced and core levels by the content (A or C) of field IPC_CLASS_LEVEL in table tls209_appln_ipc.

I ran a query in order to understand how A & C IPCs are distributed across the various patent offices (see the SQL below) expecting a 1:1 distribution.

SELECT tls201_appln.APPLN_AUTH, tls209_appln_ipc.IPC_CLASS_LEVEL,  COUNT(tls209_appln_ipc.APPLN_ID)
FROM patstat.tls209_appln_ipc
INNER JOIN patstat.tls201_appln ON tls209_appln_ipc.APPLN_ID = tls201_appln.APPLN_ID
GROUP BY tls201_appln.APPLN_AUTH, tls209_appln_ipc.IPC_CLASS_LEVEL
ORDER BY tls201_appln.APPLN_AUTH, tls209_appln_ipc.IPC_CLASS_LEVEL;

The full results of the query may be downloaded here; bytheway I publish here a graph of patenting office with more than 850.000 C and A IPCs: in red you may find the count of CORE ipcs, in blu the ADVANCED ones.

At a glance we notice that, apart from Slovenia Patent office (SI), everywhere the figure of advanced IPCs is the higher, average above 10%.
The peak of difference is to be found in some african countries (34% Kenia, 36% Malawi) but also in Iceland advanced ipcs are 31% more than core ones. Same trend in US (11%), EP (15%) and  JP (12%).