ONLINE APPENDIX to Deadlines, Work Flows, Task Sorting and Work Quality This online appendix is meant to be a companion to “Deadlines, Work Flows, Task Sorting and Work Quality.” It provides the following information: Section I: Illustrative example and model proofs Section II: Additional details on the calibration exercise Section III: Patent application process and glossary of select terms Section IV: Robustness checks Section V: Survey description, analysis of responses and summary tables (S1-S4). Notes: Throughout, D1, D3, and D5 refer to a dummy variable indicating last one, three and five working days of a month, respectively. For aggregated day-level regressions corresponding to Table 1 in the main text, these dummies refer to calendar days of a month. Some analyses are performed using patent data only because the application data did not contain the required information. In such cases, measures that rely completely on applications (e.g., application success) are not presented. All coefficients and standard errors have been multiplied by 1000 for presentation purposes. Unless stated otherwise, standard errors are clustered by firm. **: p<0.01, *: p<0.05, +: p<0.1. 1
123
Embed
ONLINE APPENDIX Deadlines, Work Flows, Task Sorting and ...webuser.bus.umich.edu/jagadees/papers/deadlines... · are costly to correct, and the inventing firm may take note of it,
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
ONLINE APPENDIX
to Deadlines, Work Flows, Task Sorting and Work Quality
This online appendix is meant to be a companion to “Deadlines, Work Flows, Task Sorting and
Work Quality.” It provides the following information:
Section I: Illustrative example and model proofs
Section II: Additional details on the calibration exercise
Section III: Patent application process and glossary of select terms
Section IV: Robustness checks
Section V: Survey description, analysis of responses and summary tables (S1-S4).
Notes:
Throughout, D1, D3, and D5 refer to a dummy variable indicating last one, three and five
working days of a month, respectively. For aggregated day-level regressions corresponding to
Table 1 in the main text, these dummies refer to calendar days of a month.
Some analyses are performed using patent data only because the application data did not
contain the required information. In such cases, measures that rely completely on applications
(e.g., application success) are not presented.
All coefficients and standard errors have been multiplied by 1000 for presentation purposes.
Unless stated otherwise, standard errors are clustered by firm. **: p<0.01, *: p<0.05, +: p<0.1.
1
I. ILLUSTRATIVE EXAMPLE AND MODEL PROOFS
2
Illustrative Example
This stylized example depicts a hypothetical patent application at a typical medium/large sized
inventing firm using an external law firm for filing. It is intended to provide a better idea of the
underlying model intuition and the filing process that our model reflects. The intuition is
similar for filings using internal attorneys.
Suppose an inventor, Curie, invents something patentable on, say, January 15. She then writes
up a 1-2 page disclosure note and forwards it to an internal review committee. The committee
may meet once a month, say January 25. Then, if the committee decides to file a patent
application, in most cases an external law firm attorney, say, John, is hired based on subject
matter expertise. As the “agent” executing the work, John has two principals: the partner
(Paula) he reports to in his law firm, and the inventing firm’s head of R&D (Peter) that John’s
firm works for (see the structure in the Figure below).
Upon receiving the work, the attorney, John, estimates the likely date of completion. He knows
that he has to interact with Curie, the inventor several times over a period of a few days to a few
weeks, to draft the patent application. John estimates his “optimal” time based on his own cost-
benefit tradeoff. Among others, John’s costs include the opportunity cost of his time and the
cost to him of making any errors in the application. Suppose John estimates that working at his
“normal” or “optimal” pace, the work will be completed and the application filed on March 20.
Then, a month-end deadline on March 31 would be immaterial to him, and a deadline on
February 28 would likely be too soon for him to complete the task (without making extremely
costly errors). In this case, he completes the work on March 20, and all of the work is billed to
the inventing firm (about $8,000 to $12,000 per application) at the end of the billing cycle,
typically March 31.
Instead of being completed on March 20, suppose John figures that some work on the
application will still be pending in the last week of March, so that at his “normal” or “optimal”
pace, John would file the application on April 3. To deal with cases like these, both his
principals have motives to impose a deadline-related penalty or incentive around March 31,
even though they know that it may cause additional errors because John has to work faster than his
optimal rate to complete by that date. For Paula, getting the work completed before March 31
would help reduce working capital costs by getting the billing of about $10,000 done a month
earlier than otherwise (if the work is completed on April 3, it would typically be billed at the
end of April). As long as the cost of the additional errors is smaller than the savings from earlier
billing, it is optimal for Paula to impose a deadline. Similarly, Peter, as head of the R&D
department, typically sets monthly/quarterly targets for patent filings, and also has an
obligation to accurately report patent-related expenditure for the month/quarter to the
accounting department. He may also have to report his progress on the invention at the next
review committee meeting. For these reasons, Peter has an incentive to try and get the
application work completed by March 31, if possible. In each of these cases, the optimal
3
completion date for the principals is different from the agent’s, and the role of the deadline is to
align the completion dates. The misalignment arises because the principals and the agents do
not share the same costs and benefits with respect to completion time.
John knows that he will likely face some pressure from Paula and Peter around the deadline.
The “pressure” on John could be from any number of mechanisms including a negative
performance review, reduction in the number of unbilled hours, potential loss of future jobs
from the inventing firm, and peer disapproval. John is also aware that he is likely to make more
mistakes or may have to file a less complete application if he accelerates his work. These can
sometimes be corrected later (e.g., by filing an additional document at a later date for a fee) but
sometimes they may not be (e.g., if some relevant matter was not included). Nonetheless, they
are costly to correct, and the inventing firm may take note of it, especially if there is a repeating
pattern of mistakes. Given these trade-offs, John may find it feasible, given the pressure from
Paula and Peter, to accelerate work to finish the application by March 31, fully knowing that the
application is likely to contain more errors, than if he had completed by April 3.
In this example, if the work had actually started earlier and had taken longer because it was a
technically complex invention or an invention with many claims, then the associated unbilled
hours (and expenses) would be larger, likely increasing the pressure from Paula (and Peter) on
John.
Note that the reason for the work being incomplete in the last week of March is immaterial (for
the predictions relating to work clustering, task sorting and work quality). For instance, it could
be that John is a procrastinator. In this case, he sub-optimally (to the principals) delays work on
the application, so that by March 20, he has two weeks of work left. John could also be an
overcautious worker who wants to reduce the probability of errors more than the principals
think is optimal, and hence ends up on March 20 with about two weeks of work left. From the
perspective of the two principals (his partner Paula, and the inventing firm R&D manager
Peter) working at an optimal rate, John should be done with the work and file the application
by March 31. Then faced with pressure from his partner (on unbilled balances) and from the
inventing firm manager (to meet the quarterly target), John may rush to file by March 31. In the
first case, the patent was procrastinated on but still ends up being rushed to be filed by March
31—hence procrastination and “rushing” are not mutually exclusive outcomes. Thus, an
application accelerated to be filed on March 31 in our framework is the result of an agent
working sub-optimally long relative to what the principals want; whether this is because the
agent was being diligent but too cautious or procrastinating is immaterial to our model.
4
Principal Agent Relationships in the Filing Process (Using External Law Firms)
Principal Agent Relationships in the Filing Process (Using Internal Attorneys)
External Law Firm Partners
[Agent for Inventing firm, Principal for Attorney – Working
z∗ represents the upper bound beyond which it is not optimal to accelerate tasks of a given complexity.
(ii) z∗ is increasing in γ for any given t∗: Taking the partial derivative of z∗ w.r.t. γ, we get∂z∗
∂γ = µ(t∗)2
2γ(1+t∗µ)2 > 0.
(iii) z∗
t∗ is increasing in γ for any given t∗: It follows from equation (1) that z∗
t∗ =[
t∗µ1+t∗µ
]. Taking the
partial derivative of this w.r.t. γ, we get ∂(z∗/t∗)∂γ = t∗µ
2γ(1+t∗µ)2 > 0.
(iv) If f(t∗) is non-increasing in t∗, z∗ is increasing in t∗ for any given γ: Taking the partial derivative of
z∗ w.r.t. t∗, we get ∂z∗
∂t∗ = (t∗)2
(1+t∗µ)2
(µ(2 + µt∗) − t∗
2µ
f(t∗)∂f(t∗)∂t∗
)> 0 if ∂f(t∗)
∂t∗ ≤ 0.
(v) Sign of[∂2(z∗/t∗)∂t∗∂γ
]: Using the result from above for the first partial derivative w.r.t. γ, and then
taking its partial derivative w.r.t. t∗, we get[∂2(z∗/t∗)∂t∗∂γ
]= 1
2γ∂φ∂t∗
1−φ1+φ where φ = t∗µ. Therefore, at
very low values of γ, (1 − φ) > 0 so that the sign of the second derivative is the same as the sign of[∂φ∂t∗
]. Similarly, for high enough γ, (1−φ) < 0 so that the sign of the second derivative is the opposite
of[∂φ∂t∗
]. The result follows from the fact the partial derivative of
[z∗
t∗
]w.r.t. t∗ has the same sign as[
∂φ∂t∗
].
I.B Proposition 3: Clustering and Complexity
Consider the derivative of[z∗
t∗
]w.r.t. t∗. The sign of the derivative depends on the sign on
t∗[∂µ∂t∗
]+ µ. Substituting µ =
√2γf(t∗) , and simplifying, we get
√2γf(t∗)
[1 − t∗f ′(t∗)
2f(t∗)
]. Thus, when
f ′(t∗) < 2f(t∗)t∗ for all t∗,
[z∗
t∗
]is increasing in t∗. Then among accelerated tasks, the mean complexity
=
∫ TTt∗ z∗
t∗ τ(t∗)dt∗∫ TT
z∗
t∗ τ(t∗)dt∗>∫ TTt∗τ(t∗)dt∗ = the mean complexity for all other days.
6
II. ADDITIONAL DETAILS ON THE CALIBRATION EXERCISE
7
II Additional Details on the Calibration Exercise
Overall Model Structure: To calibrate, we impose additional structure on the baseline model
provided in the Appendix of the paper. We assume a(x) = c(x) = x2, and b(x) = x3, so that t∗ = x
and f(t∗) = t∗, b(x)a(x) and a(x)2
b(x) are strictly increasing in x and f ′(t∗) = 1 < 2 = 2f(t∗)t∗ for all t∗. We then
simulate data for 54 months of 22 working days each for 50 firms, with patent filing jobs arriving at a rate
of one per day. To abstract from initial transitional dynamics, we drop the first six months. We set T = 25
and T = 126 so that the range of durations extends from about one month to about five months; this is
consistent with the range of days indicated by the lawyers surveyed (see Online Appendix Table S2B). We
add a stochastic component to the filing date, to capture unmodeled factors (e.g., disruptions due to health
issues or personal vacations); we allow for each application to finish up to five days before or after a scheduled
date, with a 5% probability for each of the ten alternative days.
Calibration of Standardized Deadline Penalty: Using these simulated data, we first calibrate
γ to a value that yields a last-day share equal to that in our sample of applications. We standardize this
value of γ to be γs = 1. We then compare this case to a “No Deadline” regime, where there is no month-
end penalty. As can be seen from the figure and table below, these simulated data replicate key empirical
results fairly well. In particular, we find that in the deadline regime: (i) there is significant clustering at the
month-end, (ii) complexity is higher at the month-end, and (iii) the error rate is higher at the month-end,
both conditional and unconditional on complexity.
Calibration of Error Rates: To calibrate error rates, we need to form a mapping from the excess error
for the last day of the deadline period in the simulation (defined as the mean error rate for last day less the
mean error rate for other days), to the observed excess error in the data. We begin with the “Application
Incomplete” measure. We use our real data on the top 200 patenting firms to obtain a mapping of the
additional period-end error rate in the simulations to the actual error rates on this measure. Specifically, we
form a smoothed expected excess error rate variable using a locally weighted smoothing (lowess) regression
(default bandwidth of 0.8) of the mean actual excess error rate on the simulated excess error rate.
To calibrate excess period-end error rates for each of the other seven process measures, we use the ratios
of the month-end excess frequencies for the month-end (D1) as reflected in Table 3. For instance, for the
“Separate Inventor Oaths”, the ratio of its D1 coefficient to D1 coefficient for “Application Incomplete”
(col 2 of Table 3) as 1.062. Then, we form the estimate for excess period-end errors for “Separate Inventor
Oaths” as the product of this ratio and the calibrated excess error rate for “Application Incomplete”.
Calibration of Additional Error Costs: We obtained USPTO fees associated with each of the
eight work process quality measures in Table 3. To obtain estimates of additional time required to address
the office actions associated with the process quality measures, we undertook a survey (detailed in Online
Appendix Section V.S4).1 We then used an opportunity cost per hour of $250 (based on an estimate
provided by a patent attorney interviewee), to translate the time requirements into dollar costs. These
estimates of additional costs are presented in Online Appendix Table B29. The sum over all the work
quality/error measures of the excess error rates multiplied by the total cost for each type of error yields the
total additional error costs for the period-end.
1The survey-based averages were similar to those obtained through an interview with a practicing attorney.
8
Simulation – Month-end Clustering, Complexity and Work Quality
The first row corresponds to a deadline punishment parameter s = 1, and the second row of figures corresponds tosimulation without a penalty for crossing the month-end threshold. In the second and third figures of each row, theline denotes predicted fractional polynomial fit; the area denotes 95% confidence interval; the horizontal axis indicatesworking day of the month. The dependent variable in the first figure in each row is the day-share of total patents, inthe second figure it is complexity (x in the model), and in the third figure it is error rate (as defined in the model).A month is defined as 22 working days.
.002
.003
.004
.005
.006
Day
sh
are
of
tota
l
22 66 110 154 198 242Year−day
6570
7580
Co
mp
lex
ity
0 5 10 15 20Working day
3234
3638
40E
rro
r ra
te
0 5 10 15 20Working day
Deadline (std. penalty=1), Period = 22 days
.003
.003
5.0
04.0
045
Day
sh
are
of
tota
l
22 66 110 154 198 242Year−day
7474
.575
75.5
7676
.5C
om
ple
xit
y
0 5 10 15 20Working day
3737
.538
38.5
Err
or
rate
0 5 10 15 20Working day
No Deadline (std. penalty=0), Period = 22 days
9
Sim
ulation
Regre
ssions–DayShare
,Complexityand
ErrorRate
Colu
mns
1-4
corr
esp
ond
tosi
mula
tion
wit
hout
ap
enalt
yfo
rcr
oss
ing
the
month
-end
thre
shold
,and
colu
mns
5-8
corr
esp
ond
toa
dea
dline
punis
hm
ent
para
met
er
γ=
1.
The
dep
enden
tva
riable
inC
olu
mn
1and
5is
the
day
-share
of
tota
lnum
ber
of
pate
nt
applica
tions.
The
dep
enden
tva
riable
inC
olu
mns
2and
5is
the
com
ple
xit
ypara
met
er(x
inth
em
odel
).T
he
dep
enden
tva
riable
inC
olu
mns
3,
4,
7,
and
8is
the
erro
rra
te.
Num
ber
of
obse
rvati
ons
is264
(corr
esp
ondin
gto
264
work
ing
day
sof
the
ever
ysi
mula
ted
yea
r)in
colu
mns
1and
5,
and
52,8
00
for
the
rest
of
the
colu
mns
(one
pate
nt
aday
for
48
month
sof
22
day
sfo
r50
firm
s).
No
Dea
dli
ne
(γs
=0)
Dea
dli
ne
(γs
=1)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Dep
.V
ar.
Day
Sh
are
Com
ple
xit
yE
rror
rate
Err
orra
teD
a ySh
are
Com
ple
xit
yE
rror
rate
Err
or
rate
Las
tW
.D
ayof
Mon
th-0
.000
0-0
.094
7-0
.027
60.
0192
0.00
17**
7.3
86**
4.0
72**
0.3
903**
(0.0
001)
(0.6
057)
(0.3
132)
(0.4
386)
(0.0
001)
(0.4
898)
(0.2
514)
(0.3
024)
R2
0.00
00.
023
0.02
20.
974
0.79
40.0
38
0.0
45
0.9
73
Com
ple
xit
yF
EN
oN
oN
oY
esN
oN
oN
oY
es
Note
s:R
ob
ust
stan
dard
erro
rscl
ust
ered
by
firm
inp
are
nth
eses
;**
p<
0.0
1,
*p<
0.0
5,
+p<
0.1
;
10
III. PATENT APPLICATION PROCESS
& GLOSSARY OF SELECT TERMS
11
III. Patent Application Process and Glossary
GLOSSARY OF SELECT TERMS
Term Brief Non-Technical Description$
Priority Date Date which decides what prior art affects the patentability of an application. For
many applications, this is the date on which the patent application is filed with
the USPTO.
Specification A written description of an invention that forms the basis for the claims covered
by the patent.
Claims Specific descriptions of what is protected by the patent.
Provisional
Application
An application that allows the applicant to establish an early priority date by
including only a specification and drawing (if needed) at a lower cost than a
regular non-provisional application. No claims are needed. The applicant must
file a corresponding non-provisional patent application within one year to
benefit from the earlier filing of the provisional application. The patent term
usually starts from the filing date of the non-provisional application.
Continuing Patent
Application
(Continuation)
An application that requests additional claims to an invention explained in an
earlier application that has not yet been issued or abandoned. It has the priority
date of the original application (and hence, the patent term starts from the original
priority date). No new matter may be added in this application. Hence, the new
claims must be supported by the specification in the original application.
Continuation-in-
Part
An application that partly claims the priority date of another application but
adds new matter not disclosed in that application. The original specification
must be substantially repeated in this application. Claims based on the new
matter do not benefit from the original priority date. The patent term is
determined by the filing date of the original application.
Divisional Patent An application that claims a distinct invention carved out of a pending
application, and claims the same priority date as that application (and hence, the
patent term is determined by that date). No new matter may be added. Often
results from a "restriction requirement" issued by the USPTO, because the
original application had multiple inventions (a patent can only claim a single
invention).
Non-Final Rejection Typically the first statement from the examiner rejecting the patentability of
some or all the claims. In response, applicants may argue the examiner is
incorrect, and/or amend the claims to overcome the examiner’s rejections
without additional fees. A vast majority of patents receive this notice.
Information
Disclosure
Statement (IDS)
A statement, required of all applicants, disclosing prior art relevant to the
patentability of an invention. Applicants may not knowingly omit relevant prior
art in their application.
Restriction
Requirement
Statement by the examiner that an application contains multiple inventions and
requiring the applicant to choose one of them. The other inventions are
withdrawn from the application (and may be filed again as a divisional
application).
$: This table provides a non-technical glossary for readers unfamiliar with some terms used in the paper. For
precise legal definitions, refer to the USPTO MPEP.
12
III. Patent Application Process and Glossary
PATENT APPLICATION PROCESS
In this appendix, we describe the general process of patent application. In particular, we focus
on the factors that may affect the timing of patent application and discuss how routines and
deadlines, both internal and external, may play a role in this process. We first discuss various
rules that regulate the timing of patent application. We then provide detailed steps of the
patenting process, particularly from the inception of ideas to patent filing. The specifics of the
patenting procedure vary across jurisdictions; except wherein applicable, we limit our
discussion to the case of the U.S., which is the context of our empirical investigation.
DATE OF PATENT APPLICATION
A patent is a government-granted monopoly right to an invention. Given that two inventors
may generate similar ideas independently, it is critical to determine who gets the grant of a
patent. Two systems exist for this criterion: first-to-file (FTF) system and first-to-invent (FTI)
system. As of June 30, 2015, all countries with a patent system adopt the FTF system that assigns
the right to whoever first applies for a protection of the given invention by filing a patent
application. Prior to March 6, 2013, the U.S. maintained the FTI system that assigns the right to
the person who conceived the invention first, though the person may not have filed it first. Even
in this case, however, the first person to file an application retains the prima facie right to the
grant of a patent. If a later applicant wants to claim the priority for the same invention, the
person can institute inference proceedings to determine the first inventor. But this procedure “is
costly and often very protracted; frequently it moves from a USPTO administrative proceeding
to full court litigation” (Merrill, Levin and Myers, 20041, pp 124-125). Also, in practice, FTF has
been the basis for an overwhelming majority of applications in the U.S., with less than 0.1% of
the cases ending up in inference proceedings because a second filer claims to be the first
inventor (Merrill, Levin and Myers, 2004, pp 125). Hence, even under the FTI system, the
inventor had the incentive to file an application as soon as possible to establish effective
priority. The U.S. has switched to a FTF system on March 6, 2013 with the enactment of the
America Invents Act.
The filing date of a patent application is the date on which the application is submitted for an
examination. For mailed filings, the postmarked date is the date of application. For electronic
filings, it is the date of electronic submission. In principle, the filing of a patent application
establishes the right of priority and the filing date of the first patent application becomes the
priority date for the application. This priority date determines what prior art affects the
patentability of an application. All inventions (whether already in a patent or in a pending
application) disclosed prior to that date will be considered prior art and cannot be claimed on
the application. This priority date is also important for other reasons. Based on this priority, the
claimant can file a subsequent application in another jurisdiction for the same invention. To use
1 Merrill, S. A., Levin, R. A., and Myers, M. B., (Eds.). 2004. A patent system for the 21st century. Committee on Intellectual Property Rights in the Knowledge-Based Economy, National Research Council. Washington, DC: National Academies Press.
13
III. Patent Application Process and Glossary
the right of priority, the applicant must claim the priority of the first application in subsequent
applications.2
The priority right is effective for twelve months (six months for industrial designs and
trademarks). That is, once the inventor has established the priority right, he or she has a full
year to seek protection of the invention in other countries without her priority interfered with
due to differences in legal systems.
No New Matter Rule
37 CFR 1.121 (f) says: "No new matter. No amendment may introduce new matter into the disclosure of
an application." Simply put, this rule means that an applicant cannot add any new material to a
specification (that is, the description of an invention) that was not included in the original
application.3 Any such new matter will either be rejected or can be inserted only with the loss of
the original priority date.
Note that this does not mean that no new claims can be modified or added. In fact, there is no
restriction on adding new claims or modifying them as long as the specification enables those
claims. For example, if the claims relate to “fruit” in the filing but the specifications describe
only bananas, then oranges cannot be added without losing the original filing date. Then, the
only choices will be to go with a narrower claim (only for bananas) or lose the filing date by
subsequently filing a new application which adequately describes and enables the claims to
“fruit” (which new application will have a later filing date).
Hence, as one our respondents said, “complete, broad disclosure in the first stage that can
enable the claims is the first priority for lawyers.”
NON-STANDARD APPLICATIONS
For many applications, the priority date is the date on which the patent application is filed with
the USPTO. However, there are several rules that complicate the pendency of priority right and
inventors sometimes exploit these rules strategically to preempt others, reduce filing costs or
effectively extend the period of priority right. These rules allow for non-standard applications
which can influence the inventing firm’s decision on the actual timing of filing a patent
application. Below we discuss some of the most important ones in detail.
Provisional Applications
A provisional application is a U.S. national application for patent which allows filing without a
formal patent claim or any information disclosure statement while still establishing an early
effective filing date for a later-filed standard ("non-provisional") patent application. To file a
provisional application, the applicant has only to include a brief description of the invention
2 For instance, an inventor may have filed first in Japan and then files an application in the U.S. Then, the
applicant should include in the application document the information of the earlier Japanese application
including the priority date under the ”Foreign Application Data”. 3 Definition of specification: “The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention” 35 USC 112 (pre-AIA).
14
III. Patent Application Process and Glossary
and any drawings that help understand the invention, along with the standard information
items such as inventor names and addresses. Hence, it provides the inventor with a lower-cost
means to establish the priority right. After filing a provisional application, the inventor has a
pendency period of twelve months, which is non-extendable, to decide on one of the following:
(i) file a corresponding non-provisional application; (ii) file a grantable petition to convert the
provisional application into a non-provisional application; or (iii) simply let the pendency
expire. If the inventor files a corresponding non-provisional application, the term of the granted
patent is measured from the date of this non-provisional application, not the date of the
associated provisional application. However, if a grantable petition for a conversion is filed, the
term is measured from the date of the provisional application. Therefore, if filed appropriately,
the provisional application provides the inventor with two distinct benefits. First, it reduces the
cost of establishing the priority right, especially when the inventor faces a severe time constraint
in preparation or when the inventor wants to minimize initial investments until she is sure of
whether to ultimately seek a patent. Second, by filing a non-provisional application based on
the provisional application, the inventor can effectively extend the period of patent protection
by up to one year.
Nonetheless, a provisional application is not a panacea to meet deadlines. In general, our interviewees
were hesitant to recommend provisionals as a substitute to a fully prepared non-provisional
application except in certain strategic cases where the provisional is actually a fully prepared
application (e.g., in pharma where a provisional gives the assignee an additional year of
protection; not surprisingly, in our data drugs and medical patents account for less than 4% of
non-provisional patents but nearly 15% of provisional patents). The biggest risk to provisionals,
they stated, is the inability to add new matter to the subsequent non-provisional application
without losing the priority date. As one respondent put it, ““Provisionals can be dangerous”.
Though easier and cheaper than the complete non-provisional app, it may not be thought out carefully,
and may miss important material (e.g., material desired for protection, material necessary to enable the
invention, etc.). If the provisional is filed with these defects, it is not possible to correct the defects with a
subsequent filed application without losing the priority claim. This may become a major issue if the
inventor has already presented that omitted material publicly. It may prove to be a fatal error, for
example, if the provisional application was not enabling. Indeed, the subsequent filed application would
not be able to claim priority to the provisional application and the loss of the priority claim may result in
prior activities creating a barring event.” Indeed, Crouch (2011)4 mentions that over 40% of the
provisional applications are abandoned without being used to claim priority based on those
application dates. The same risk is associated with the use of continuations, which require that
the continuation not include anything which would constitute new matter if inserted in the
original application.
Patent Cooperation Treaty (PCT) Applications
A PCT application is a patent application filed under the PCT, an international law treaty that
provides a unified procedure for filing patent applications. The inventor may file a PCT
application as the first filing or as a subsequent filing referencing an earlier application in a
jurisdiction. If filed, a PCT application establishes a filing date in all member countries (148 as of
June 2015) and hence the inventor has no need to file separately in these countries to establish
the priority right. However, to seek a patent on the invention in a specific jurisdiction, the
applicant must enter the national phase within the specified period (typically 30 months from
the date of PCT filing or the earliest priority date) by paying the national fee and, if required,
provide a translation of the application. Filing through the PCT, instead of directly in the
countries of interest, the inventor can enjoy several benefits. First, since the inventor only needs
to file in a single language with a single jurisdiction, the PCT application significantly reduces
the cost of establishing the priority rights in multiple countries. Second, it buys time for the
inventor to gather further information before deciding where else to file patent applications.
Third, owing to the 30 month of international phase, the PCT application practically extends the
period of priority in a given jurisdiction by up to 18 months.
Continuing Applications
A continuing patent application is a patent application that follows and claims priority to an
earlier patent application (the “parent'” application) filed by the same applicant. There are three
types of continuing patent applications: (i) continuation; (ii) divisional; and (iii) continuation-in-
part.
A continuation application is filed when the applicant wants to include additional claims (but
no new subject matter) to an invention disclosed in the parent application that is still under
examination. Though potentially useful in cases where some claims in the parent application
are rejected while others are allowed, the continuation application is often used to simply buy
more examination time. In this case, the priority date of the original application applies to this
continuation. However, this application gets a shortened patent term since the patent term starts with
the filing date of the earliest parent application.
Applicants usually file a divisional application when the examiner determines that the parent
application contains more than one invention and issues a “restriction requirement”. In this
case, the applicant can decide which invention to include in the pending application and choose
to prosecute the rest as divisional applications. As with continuations, the patent term is shortened.
In a continuation-in-part application, the applicant can add subject matter that is not disclosed
in the parent application but repeat a considerable portion of the parent's specification. This
type of application is used when the inventor comes up with enhancements to the invention
after filing the parent application. In this case, the new matter does not receive the priority
benefits of the parent application but the patent term starts from the filing date of the earliest
parent application. This also takes the longest time to prepare among the three types of
continuing patents (see table below).
Although different in the usage and the associated requirements, all of three types of continuing
application claim, in full or in part, as the priority date, the filing date of the parent application.
The continuing application provides opportunities and considerable flexibility for the inventor
to claim the full scope of a disclosed invention by filing as many continuation applications as
deemed necessary and thereby seek a broadest possible protection of the invention. However,
16
III. Patent Application Process and Glossary
this procedure is subject to criticism that such flexibility creates uncertainty about the ultimate
scope of a given patent application.5
As with provisionals, continuing patents are not a panacea to meet deadlines. For instance, one may
be tempted to file a not-so-well prepared application assuming that any problems can be fixed
later using continuing applications. First of all, they are each treated as a new application with
regard to USPTO fees, thus considerably increasing the costs. Moreover, except continuations-
in-part, the other options do not allow any addition of new matter, thus allowing only minor
changes to the original application, if any. Perhaps, most important, they all result in loss of
patent term for the new application.
The loss of patent term is particularly important for continuations-in-part. It requires the
inventors to seriously examine why they want to file a continuation-in-part on an improvement
that the applicant believes is novel and nonobvious over the previously filed subject matter
since they will be giving up patent term, even if all claims include the new matter. With a
continuation-in-part, there may be little benefit with a guaranteed loss of patent term.
Priority Date Patent Term Effort to Prepare$
Continuation Same as Earliest
Parent Application
Starts from Filing
Date of Earliest Parent
Application
Small; 10-20% of
initial filing; 1-3
hours
Continuation-
in-part (CIP)
Same as Earliest
Parent Application
for “old” matter
Filing date of CIP
for new matter
Starts from Filing
Date of Earliest Parent
Application
Medium-Large; 20-
50% of initial filing;
10-30 hours
Divisional Same as Earliest
Parent Application
Starts from Filing
Date of Earliest Parent
Application
Small; 5-10% of
initial filing; 1-3
hours
$: Based on estimates provided by two patent attorneys we interviewed.
PATENT APPLICATION PROCESSES
We now turn to the detailed steps of patenting process. Depending on the inventor's status of
affiliation, the application process can be quite different across inventors. A patent can be
legally assigned to an individual or to an organization. In the former case, it is the inventor who
owns the claim to the invention and, if any, the economic rent from it. In the latter, it is the
organization who owns the claim. In general, organizations include universities, research
centers, government bodies and corporate firms. Our discussion will focus on corporate firms.
For convenience, we call the inventor who also claims the ownership to herself as the “inventor-
5 In August 2007, the USPTO announced a new rule that limits the number of continuation applications
for each invention disclosed in the original patent application. However, in March 2009, after facing
strong opposition from many parties including lawsuits, the USPTO finally withdrew the proposed
changes to the continuation procedure.
17
III. Patent Application Process and Glossary
owner”, and the inventor whose patent is legally assigned to an organization as the “R&D
personnel”.
Patent Application Process for Individuals
For the inventor-owner, there is no typical process of application; in principle, the inventor
decides whether and when to apply for the protection of an invention.6 Also, the inventor faces
little procedural constraints in filing an application. The inventor may work with a patent
lawyer for the actual filing but key decisions lie with the inventor.
Patent Application Process for Corporate Firms
For R&D personnel at corporate firms, the patent application process is more complicated. The
process begins with the inventor's disclosure of an invention and ends with either the
abandonment of the idea or the filing of a patent application. Various steps and feedback loops
exist in between.
The figure below illustrates the process flow for a typical firm that has established procedures
for patent prosecution.7 When the R&D personnel come up with a patentable idea, she first
submits it to an internal review committee for an assessment. This “disclosure” document is not
yet a full description of the idea; it could be a one-page memo or a short technical note that
outlines the subject matter of the invention. The review committee typically comprises senior
managers in the R&D department and the legal team that is responsible for patent prosecutions.
The committee regularly meets to discuss and determine if ideas submitted from various
research teams are worth pursuing for a patent.8 If an idea is approved by the committee, it is
then drafted into a formal patent application document. The inventor may write the first draft
and pass it on to the legal staff of the inventing firm (i.e., internal lawyer), or the legal staff may
write up the draft based on the information the inventor provides. In either case, multiple
rounds of meetings, online and offline, take place between the inventor and the lawyer before
the application document is finalized. Many firms entirely outsource this portion of the patent
prosecution to external law firms (i.e., external lawyer), while some use a mix between internal
and external lawyers. In our data, about 79% of the applications were represented by external
lawyers.
6 In the U.S., 14.4% of all patents granted during 1976-2009 were assigned to individual inventors. 7 Note that while this flow is broadly representative of the process followed by the firms we spoke with,
specific practices vary across firms. For instance, the ranking system and the disclosure review committee
composition likely varies across firms. 8 This step of assessment often involves more than a simple selection/rejection. For those ideas that are
selected for the next step, the committee could further sort them into several “grade” levels depending on
the expected value of the idea. The idea that receives a higher grade gets a higher priority in prosecution
(e.g., expedited processing). Lower-graded ideas may be pursued for patenting in a limited number of
jurisdictions or even only in a single market. The rejected ideas are not always bad ones. For firms that
are cutting-edge in certain technological fields, even the rejected ideas can be valuable from other
perspectives. One of our interviewees recalled incidences in which some academics spent their sabbatical
year at his firm to dig through past rejected disclosures and turned some of them into academic papers.
Also, we were told that some ideas are not approved for patenting, not because they are less than worth
patenting, but because they are assessed to be more valuable if kept secret.
18
III. Patent Application Process and Glossary
Once the patent application is finalized, the lawyer (either internal or external) files it with the
patent office in the corresponding jurisdiction (e.g., USPTO, WPO). In the U.S., most filings
today are done through the electronic filing system (EFS), which was introduced in March 2006.
Before the introduction of the EFS, USPS express mail was the primary method of filing an
application. The date stamped by the post office on the mail is considered as the filing date by
the patent office. The patent office gives a unique identifier to each application and assigns it to
a patent examiner for a review.9 During the examination process, the patent examiner may issue
office actions that demand follow-up actions from the applicant such as further clarification on
the invention's specifications. These requests are mostly handled by the lawyers who drafted
the application, but sometimes with input from the inventor. For many firms, external lawyers
are also responsible for the maintenance of granted patents such as alerting the client to renewal
notices and, upon the client's renewal decision, paying renewal fees on behalf of the client.
Deadlines and Related Incentives for Agents
As in many other tasks, deadlines play an important role in the process of patent application.
The types of deadlines and the associated costs of missing (or equivalently, incentives to meet)
9 The actions taking place within the patent office critically impact a number of specifics of a patent,
including the breadth of coverage, references, and of course, granting and the timing thereof. However,
given our primary interest in the timing of application, we abstract away from the internal process of the
patent office.
19
III. Patent Application Process and Glossary
these deadlines differ depending on the stage of application process and the agents involved in
the process. For instance, for internal management purposes the inventing firm may place
certain periodic target dates by which R&D personnel are required to disclose patentable
inventions. Also, internal review committees described earlier may be convened only
infrequently (e.g., monthly, quarterly) to minimize task interruptions or better coordinate across
different functions involved in R&D. Hence, these meetings may also entail certain deadlines
for collecting ideas to be evaluated. Our interviews and surveys of inventors as well as legal
staff at inventing firms, presented in this Online Appendix V.S1, confirm the practices of these
internal deadlines.
More importantly for our study, agents that work on documenting and filing the applications
(i.e., internal lawyers for the internally filed applications, and attorneys at external law firms for
outsourced applications) face different types of deadlines and associated costs and incentives.
Some of the deadlines are statutory, such as the 12-month period to claim the right of priority
based on a foreign application. These statutory deadlines apply equally to internal lawyers and
external lawyers. Our interviews with practicing lawyers suggest that missing statutory
deadlines, though not common, can have severe consequences for the lawyers as it often leads
to a loss of priority right. For external lawyers, it may very well mean losing the client or (in
severe cases) even a malpractice lawsuit.
More traditional nonstatutory deadlines are widely prevalent in the filing process. Our
interviews and surveys, part of which are presented in this Online Appendix V.S2 and V.S3,
indicate that internal lawyers tend to work with deadlines arising from work schedules in R&D
departments of the inventing firms. The most common cost of missing deadlines for internal
lawyers, according to our surveys and interviews, was lower performance evaluation,
especially if that was a pattern of behavior. One interviewee noted a practice of monthly status
reports in one prominent patenting firm that listed completed and remaining work on
applications for each of the internal attorneys, which could serve as informal deadlines (as it is
reviewed by superiors and peers), even when there are no formal penalties or incentives
associated with it. In outsourced applications, clients frequently impose deadlines to external
lawyers to complete the task (e.g., file an application by a certain date). When the client-law
firm transactions are based on a renewable fixed-term contract (typically one year), both parties
establish a routine for settling the account, which includes regular billing cycles (typically
monthly). Under this type of arrangement, external law firms are paid only after filing of the
application. Hence, at least for partners at law firms, these billing cycles function as deadlines in
managing the flow of application work. When missed, these nonstatutory deadlines can also
have important consequences for the partners and lawyers. Indeed, our interviewees and
survey respondents suggested as possible consequences the loss of monetary incentives, lower
performance evaluations, and pressure from colleagues. For partners at external law firms,
unfiled applications at the end of the billing cycle would mean unbilled hours for the work that
has already occurred. Missing deadlines, especially if repeated, could also lead to loss of the
client.
Our interviewees indicated that rework costs are often passed on to the inventing firm by the
external law firm. However, our interviewees suggest that this specific agency problem is
mitigated by the law firm’s consideration of potential consequences (such as bad performance
reviews and client dissatisfaction) of too frequent or too obvious errors. In fact, some of our
20
III. Patent Application Process and Glossary
interviewees suggested that large clients, working with multiple law firms for patent
prosecution, regularly review the quality of the filing work (in terms of timeliness,
documentation errors, etc.) and determine the assignment of volume in the subsequent period,
and sometimes even demand removal of lawyers whose performance falls below certain
threshold. Thus, even in cases where costs are passed on, there are indeed long-term
consequences for external law firms.
In this setting, though the principal may be unable to perfectly monitor the agent’s efforts, (s)he
can still observe ex-post work outcomes that will help the principal make inferences about the
agent’s efforts. For instance, one of our interviewees said “[c]lients also have the opportunity to review the quality of the patent application after the application is filed and when the Patent Office sends a response to the filed application. For instance, the client may recognize that a patent was poorly drafted if the examiner points out errors that occurred when drafting the application.”
Drivers of Duration of Patent Application Preparation
In our formal model, the defining characteristic of a patent application is a scalar that we term
“complexity”. This complexity in our model is related closely to the duration required to
prepare the application.
To confirm that this usage of complexity is consistent with practice, we requested the input of
one of our interviewees, an experienced corporate attorney (who was exceptionally generous
with his time and expertise), on the drivers of the time it takes for the agent to prepare a patent
application document.
The interviewee confirmed that complexity was indeed related to duration. To quote directly
from his email “Complexity of the invention: Some inventions are more technically complex than
others—requiring significant review to determine the best way to claim the invention.”
The interviewee also notes the role of searching through prior art in increasing complexity and
hence duration of work. In particular, he notes: “Crowded prior art: The number of prior art
references can increase complexity - particularly if they are all very close to the point of novelty that is
being claimed. Indeed, if there are a lot of close prior art references, it is necessary to determine how much
you can claim without running into references that already disclose the claimed subject matter. If there
are many desired points of novelty in the invention, each point of novelty needs to be explored to
determine the breadth of the claim that will provide desired claim scope while still defining over the prior
art.”
Another key factor is the quality of the invention disclosure by the inventor. The attorney noted:
“Sometimes experimental data needs to be provided to support claim features—sometimes the
experiments have not yet been performed—so we may need to wait for experimental results to support
some claim features.”
Lastly, duration also depends on the number of points of novelty claimed in the application.
The interviewee notes: “Some inventions only have one aspect that we are trying to claim. Others have
many new features that need to be claimed. When there are many new features, extra time needs to be
spent to fully explain and claim each feature individually as well as in combination with other
features. This can increase the number of claims and time necessary not only to describe each point of
novelty, but how each point of novelty can be used alone or in combination with the other points of
novelty.”
21
IV. ROBUSTNESS CHECKS
22
IV. Robustness Checks
In this section, we present rich evidence from our primary data and additional empirical
analyses to check the robustness of our results while eliminating several alternative accounts.
ARE THE CLUSTERING PATTERNS ATTRIBUTABLE TO DEADLINES?
Evidence From Interviews and Surveys of Practitioners
The most direct support for the observed clustering being due to deadlines comes directly from
our surveys and interviews of practitioners. When asked about the possible reasons for the
observed clustering, not a single one of the 64 respondents in our attorney surveys chose a non-
deadline related reason. Also, every one of the 16 interviewees attributed the observed
clustering to organizational deadlines.
The evidence suggests a strong role for each of the dual principal-agent relationships embedded
in the patent application process. In particular, we found important roles for billing deadlines
pressure-related to working capital considerations imposed by law firm partners (serving as a
principal for the external patent attorneys), as well as reporting and planning routines-related
deadlines imposed by the inventing firm (serving as a principal for both internal and external
attorneys).
Stressing the role of monthly billing deadlines, one interviewee said, “The most likely thing that
could be driving these results is law firm internal revenue issues…” Another attorney remarked,
“There may be a lot of pressure on attorneys at law firms to bill out any billable work for that month. So,
they may have a motivation to complete work by the month-end so they can invoice the clients for it.” The
same respondent also stressed a role for inventing firm routines, “[Also], … many inventing firms
(especially the larger, sophisticated ones), are likely to have metrics and cyclical reports (monthly,
quarterly, and yearly) to encourage employees/attorneys to file patents in a timely manner.”
Consistent with this, in our inventor survey, we find a positive correlation between the pattern
of monthly clustering and the reporting deadlines for external attorneys. Specifically, Figure
S1A and Table S1E show that organizational clustering patterns correspond exactly with the
reported deadlines (annual deadline firms show year-end, quarterly deadline firms show
quarterly, and monthly deadline firms show monthly clustering). Additional evidence comes
from our law firm attorney surveys. We compared filing patterns for firms whose respondents
cited monthly billing cycles as a possible reason for the clustering to filing patterns for firms
who did not mention these cycles as a reason. Figure S2B shows that those who mentioned
monthly billing cycles (indicated by a solid blue line) are more likely to exhibit month-end
clustering.
Our surveys and interviews suggest a prominent role also for deadlines imposed by the
inventing firm, both on external and internal lawyers. About 53% of external attorneys (35 of 65
respondents), and about 69% (9 of 13 respondents) of internal attorneys suggested a role for
inventing firm (client) imposed deadlines in the monthly clustering pattern. A number of
corporate attorney interviewees stressed the importance of quarterly, half-yearly and yearly
accounting and planning routines. An interviewee from a very large patenting firm noted that
patent filing and related expenses amount to several million dollars for top filers, and because
these expenses need to be reported in quarterly accounting statements, it becomes imperative to
plan and file patents within quarterly time windows, leading naturally to quarter-end
deadlines. Consistent with this comment, six of eight attorneys that filed patents internally
23
IV. Robustness Checks
reported that applications are some- times or often expedited to meet non-statutory deadlines
usage of monthly reporting schedules is common in R&D departments of large firms, which
could spur internal attorneys to finish up work before the month-end reporting deadline.
Our surveys also indicate that patent attorneys face significant costs if they do not meet a
deadline. Nearly three-quarters of all external attorney respondents said that missing a deadline
is likely to lead to lower performance evaluations sometimes or more often (Table S2D). More
than a third rated such a possibility as often or higher. Related qualitative comments on possible
consequences of missing a deadline also pointed to such costs (e.g., reduction in available work,
unhappy client, and payment for loss or damages). We performed a more formal analysis of whether
these perceptions of costs are associated with month-end clustering. We defined a dummy
variable costlymiss as one if the respondent rated any of the three aforementioned consequences
as sometimes or higher. We then plotted the month-end share of filings for these two groups of
patents. The results presented in Figure S2D clearly indicate a much higher prevalence of
month-end clustering among the group that rated the consequences as more likely (solid line
represents costlymiss = 1). This supports a direct association between the disincentives related to
missing deadlines and month-end clustering.10
Our respondents also provided their opinion on the negative perceptions associated with
several transactions during the patent prosecution process. We created a dummy variable,
negpercep, defined as one if the respondent rated the occurrence of any of four transactions
(application incomplete, request for extension of time, new drawings to be filed, additional
filing fees) as somewhat negative or worse. We then plotted the month-end share of filings for
these two groups of patents. The results clearly indicate a much lower prevalence of month-end
clustering among the group that had a more negative perception of receiving these office
actions (solid line represents negpercep = 1; Figure S2A). This is consistent with the respondents
being aware of the costs of rushing a filing at month-end, and avoiding it if they believe the
costs are high.
Finally, we put this all together in regressions of a month-end dummy on costlymiss, negpercep,
billcluster (defined as one if they mentioned billing cycles as a possible reason), and clientcluster
(defined as one if they mentioned client-imposed deadlines as a possible reason) with month
and application year fixed effects. Table S2G presents the results. In line with the corresponding
figure, the coefficients on costlymiss are strongly positive and statistically significant throughout.
This supports an association between the disincentives related to missing deadlines and month-
end clustering. The coefficients on negpercep are negative and generally statistically significant,
indicating that attorneys avoid rushing at month-ends if they believe the costs of rushing are
high. Billcluster is generally positive, consistent with our argument (though the significance
weakens on longer windows), while clientcluster stays insignificant throughout, suggesting that
incentives of attorneys may be a relatively more important driver of month-end clustering of
filings.
10 The results in relating baseline results to firm size (Hypothesis 4, Table 4, Section 4.4) suggest that inventing firm indeed plays an important role in the clustering patterns. In addition, results in this Online Appendix Tables B7A and B7B, discussed below, point to a big role for fiscal-year end deadlines in explaining month-end clustering patterns.
24
IV. Robustness Checks
Additional Empirical Tests on Firm Routines
In the baseline analysis in Section 4.1, we found that filings by individuals show little or no
clustering at month-ends (except before October 1, when the USPTO usually makes fee
changes). In line with this, we do not see any month-end effects on quality or complexity for
patent filings by individuals (Tables B3 and B4). We also compared large (more than 100 patents
over the data period) and small firms (only one patent over the data period), and find that large
firms exhibit significantly greater clustering compared with small firms (Table B5). The large
firm size effect was robust to the inclusion of lawyer fixed effects and lawyer volume (Table B6).
Together, these results strongly establish that larger firms are more likely to have month-end
clustering patterns than individuals or small firms.
In addition, following Oyer (1998), we exploit changes in fiscal year-ends to see if clustering
patterns within firms shift in the expected direction. Though this focuses not on month-end
clustering but on year-end clustering, it provides a useful link between a known routine-related
deadline and clustering of filings. We find that, across firms, clustering is indeed correlated
with fiscal year-ends; specifically, the share of filings that occur in the last month of the fiscal
year is significantly larger than that for any other month (Table B7A). Further, we regressed a
dummy variable that is one if the patent was applied in the month in which the old fiscal year
ended, on a dummy variable for the time period prior to the fiscal year switch and other
controls (application year, month, and firm fixed effects). The coefficient on the old fiscal-year
dummy is strongly positive, implying that the propensity to file at the old fiscal year-end
diminishes after the switch (by 3.6%, Table B7B).11 Together, these tests strongly support a role
for large firm routines in driving the observed clustering patterns in patent filings.
ARE INVENTORS THE PRIMARY SOURCE OF CLUSTERING?
Thus far, our arguments have focused on the final stages of the filing process, just before the
application is filed. However, it is possible that routines associated with upstream stages, such
as inventors’ performance evaluation cycles, are causing the clustering patterns. Based on our
surveys, for several reasons, we believe that the observed clustering is unlikely to be due to
inventor behavior.
Our inventor survey suggests that it typically takes anywhere from a week to 12 months (with a
mode of 1–3 months) between management approval and patent filing, implying that inventors
do not have control over the exact timing of the filing (Table S1A). Our surveys of attorneys
yielded a similar timeline between invention disclosure to the attorney and filing (Table S2B).
Further, our survey of attorneys confirmed that the exact timing of patent filing is primarily
determined by either the law firm attorneys (42% of the responses) or corporate legal staff
(43%); only 4% replied that inventors play any role in deciding the timing of the filing (Table
S2A).
The modest (if any) role of inventors is consistent with our inventor survey evidence that the
number of patents applied for is only one of the criteria for performance evaluation of the R&D
11 Corresponding changes in complexity and quality were noisy, though generally in the expected direction, suggesting that the magnitude of these effects don’t vary sharply across month- and fiscal-end deadlines (Table B8).
25
IV. Robustness Checks
organization, and in fact is not even the most frequent one (Table S1B, Panel A). Even for
individual R&D staff, though patent filing and grants were somewhat important, it was not the
most important criteria for internal performance reviews (Table S1C). As a direct test, we
checked for clustering patterns in the reported month of inventor performance evaluation; we
did not find evidence for any significant clustering. In fact, fewer patents are filed in the month
inventors reported they are evaluated for performance compared with other months (Table
S1D). Moreover, no inventor reported having monthly performance evaluations (Table S1F).
Deadlines related to inventor performance evaluation were not mentioned as a possible reason
for the observed clustering in any of the over 20 interviews we conducted with attorneys. In
fact, in these interviews, none of the R&D staff were aware of the clustering pattern in patent
filings. Together, these pieces of evidence strongly suggest incentives and penalties related to
inventors are unlikely to play a significant role in the observed clustering of patent filings.
IS THE MONTH-END CLUSTERING ATTRIBUTABLE TO ECONOMIZING ON
COMMON COSTS?
Firms may choose to file patents together to economize common costs of filing or other
coordination reasons unrelated to deadlines (e.g., R&D managers may choose to interact with
external attorneys about filings on a certain set of days to avoid distractions to their primary
jobs). Alternatively, firms may choose to file all patents related to the same invention on the
same day in order to avoid having different priority dates for those patents. At the outset, note
that these possibilities imply clustering, but not necessarily at month- ends. Nonetheless, we tried
estimating the magnitude, if any, of the bunching of applications due to such reasons. For all
applications filed by a firm in a given application-year-month cell, we computed the share of
applications filed on any given day. We then computed the deviation of these shares (excess rate)
from a hypothetical uniform share based on the number of applications in a given month (e.g.,
if the firm filed 10 patents in a given month uniformly (i.e., one-a-day), then the uniform daily
share will be 1/10; on the other hand, if it filed all the patents on a single date, the average daily
share will be 1). The mean excess rate in the sample was 0.0336, implying that the probability
that applications are processed in a batch is about 3.36 percentage points higher than the
expected uniform rate. The excess rate decreased sharply with the number of monthly
applications (since the applications are more likely to spread out over the month) but never
reached zero, indicating the existence of some bunching during the month, possibly due to
economization of costs related to filing. However, this excess rate was significantly higher at
month-ends (Table B9). If the clustering is purely due to economization of costs and unrelated
to deadlines, there is no reason to expect this excess bunching to be higher at month-ends.
In a different test, we used shared backward citations as an indicator of different patents
belonging to the same invention.12 We constructed all patent-to-patent pairs filed in the same
month by the same firm, and computed for each of these approximately 88 million patent pairs,
a dummy for sharing a backward citation, and the number of backward citations shared
between them. A pair of patents filed by the same firm on the same day is about 12 times more
likely to share a backward citation than a pair of patents filed by the same assignee on different
days of the same month (0.092 vs. 0.0075). But, we find that conditioning on the pair of patents
being filed on the same day, the probability that any two patents share at least one backward
12 We thank an anonymous reviewer for suggesting this test.
26
IV. Robustness Checks
citation is significantly lower at month-ends than at non-month-ends (Table B10). Likewise, the
number of shared backward citations at month-ends is also significantly lower than that at non-
month-ends. This is inconsistent with month-end clustering primarily being driven by the
incentive to economize on costs related to the same invention; if that were the case, we should
see that for the patent pairs filed on the same day, the pairs filed at month-ends share a greater
number of backward citations than that of those filed at non-month-ends.
Our theoretical framework posits clustering as arising from rushing to avoid deadline penalties.
A plausible alternative is that observed clustering is instead caused by waiting to save on
common costs by filing at periodic (incidentally at month-end) check- points. A number of
results suggest this is not the case. If waiting were causing clustering, one would expect the
decline in the rate of filing to be toward the middle or latter part of the month, when the waiting
would be for a few days, rather than at the beginning of the month, when the wait would be
close to a month. However, the results show a decline in the filing rate at the beginning of the
month (e.g., the negative coefficient on days 1–7 in columns 7–9 of Table 1). Further, we observe
clustering at the month-end even for subsamples of firms with just one filing per month, which
additionally rules out a role for fixed costs savings. Also, there appears to be no plausible
reason for higher error rates (or more-complex applications) at month-ends, if clustering were
due to waiting; in fact, delayed filing may allow for some errors to be corrected, so that we
could expect fewer errors at month-end. Another piece of evidence against this “waiting”
hypothesis comes from our analysis of accelerated filing using foreign priority dates (discussed
below). That analysis shows that applications for which the expiration date to file based on
foreign priority falls within the first few days of a month are significantly more likely to be filed
in the last five days of the previous month. Because these applications have guaranteed priority
dates based on their foreign applications, they have no compelling reason to be accelerated.
Nonetheless, the filers of these applications chose to file earlier rather than waiting until the
deadline. Hence, this analysis renders additional support for acceleration, rather than waiting,
as the driver of clustering patterns we document in this study.
IS CLUSTERING DRIVEN SOLELY BY PRIORITY DATE CONSIDERATIONS?
One of the sources for deadlines discussed in our theoretical framework (Section 2, and Figure
1) is in- congruence between the principal and the agent in perceived benefits from an earlier
priority date. The heterogeneity of results with technology cycle time (Hypothesis 4, and
Section 4.4) confirms a role for priority date considerations. A question that arises is whether all
of the observed clustering may be driven by this concern, with no role for any of the other
sources for deadlines discussed in our theory (planning, coordination, or working capital
considerations). Note that as with economizing on fixed costs, simply rushing for getting an
earlier priority date neither directly causes clustering (if all applications get accelerated) nor
explains why clustering happens exactly at month-ends.
To check if earlier priority date is the sole consideration, we examined U.S. patent applications
that claim priority to a foreign patent. Because these applications already have a priority date,
they do not have any rushing benefits related to priority date considerations. However, they do
have a statutory deadline to meet; they must be filed within twelve months from the day the
foreign application is filed. In fact, we find that these filings also show month-end clustering,
though not quite as sharp as other filings (Figure B1 and Table B11). Because such clustering
27
IV. Robustness Checks
may be driven by the timing of the original filings, we examine if such filings with a statutory
deadline on the first few days of a month are more likely to be filed 1–5 days earlier (so that
they are filed at the previous month-end). We find evidence of such acceleration, which strongly
suggests that there is rushing of tasks to meet a month-end deadline, over and above the
benefits of an earlier priority date (Table B12). Finally, we checked and found that the month-
end results for higher complexity and lower work quality generally hold for patents with
foreign priority dates (Table B11). These results confirm that early priority dates are not the sole
driver of month-end clustering patterns in the data.
THE ROLE OF SPECIAL APPLICATIONS (PROVISIONALS AND CONTINUING
PATENTS)
In our analysis so far, we used data on all applications. However, some of these applications are
different from the standard non-provisional application, and there may be a concern that these
applications are somehow biasing or driving the results. Our results in this section strongly rule
that possibility out. There are four main categories of special applications: (i) provisional, (ii)
continuation, (iii) divisional, and (iv) continuation-in-part (explained in detail in Section III of
this Online Appendix).
First, we checked and verified that the significant month-end clustering, higher month-end
complexity and lower month-end work quality (except for the examiner-added cites measure)
results hold in the sample excluding these patents, with little change in the magnitude of effects
(Table B13).
Second, we examined if clustering patterns for these types conform to what we would expect
given the key sources of deadline pressure. Note that in our framework, deadline effects are
more likely if applications are: (i) time/effort-intensive (hence, higher fee, which then is likely to
generate implications for planning/coordination or working capital), and/or (ii) priority-
relevant (so that firms may need task completion using a deadline earlier than the attorney
would otherwise want to). Continuation applications are modifications of previously filed
patents, and the application cannot contain “new matter” (USPTO MPEP, Section 201.06).
Divisional applications are “for a distinct or independent invention, carved out of a pending
application and disclosing and claiming only subject matter disclosed in the earlier or parent
application” and “often filed as a result of a restriction requirement made by the examiner”
(USPTO MPEP, Section 201.06). Thus, the time and effort intensity is limited for both (as
confirmed by two practitioners; see “Patent Application Process and Glossary” in this Online
Appendix III). Further, both these types of applications take the priority date from the original
or parent patent application, so the timing of filing of neither type of applications is priority-
relevant. Thus, these two types of patents are likely to face little, if any, deadline pressure.
Provisional applications are much shorter than standard non-provisional applications
(minimally one cover page plus one or more pages; Quinn 2013), so deadline pressure from
time/effort-intensity is likely to be low.13 But these are relevant for setting priority date – in fact,
these are primarily used to secure a priority.14 In contrast, continuation-in-part applications take
13 Quinn, G. 2013. The benefits of a provisional patent application (blog). www.ipwatchdog.com. Updated September 14, 2013. 14 Claims or oaths are not needed for provisional patents, and these are not examined for patentability (per USPTO MPEP, Section201.04, http://www.uspto.gov/web/offices/pac/mpep/s201.html).
Mechanical, and Miscellaneous). We confirmed that the month-end clustering pattern is stark
for all the categories (Table B17A).17 Sorting and work quality results are largely robust as well
(Panels A to F of Table B17B). These results confirm that the baseline results are quite broad-
based and not influenced by one particular technology.
(v) Excluding provisionals, continuations, continuations-in-part, divisionals, and patents with
foreign priority more than doubled the magnitude of the observed clustering and work quality
effects, particularly on work process (Table B19).
(vi) The results were largely robust to using alternative time windows to measure citations and
renewal probabilities (Table B20A), and to excluding self-citations (Table B20B).
15 A related question is why might there be no additional clustering of provisional applications since they provide a cheap way to obtain earlier priority. When faced with a deadline, one would expect firms to file a provisional application. We find no support for this view. The coefficient on the provisional dummy in Table B14A is strongly negative for two of the three month-end dummies, and only weakly positive for D1; so, they are not used disproportionately at month-ends. As explained in Section III of this Online Appendix, a provisional application is not a panacea to meet deadlines. In general, our interviewees were hesitant to recommend provisionals as a substitute to a fully prepared non-provisional application except in certain strategic cases where the provisional is actually a fully prepared application (e.g., in pharma where a provisional gives the assignee an additional year of protection). The biggest risk to provisionals, they stated, is the inability to add “new matter” to the subsequent non-provisional application without losing the priority date (consistent with the discussion in Quinn 2013). 16 We thank anonymous reviewers and the Associate Editor for suggesting some of these robustness checks. 17 Clustering is stronger for Computers & Communications, and Electrical and Electronics (presumably because of stronger competition in these categories), and somewhat lower in Drugs & Medical, relative to the baseline “Miscellaneous category.”
29
IV. Robustness Checks
(vii) Using all available data (instead of 2001–04) to analyze examiner citations yielded stronger
results (Table B21).
(viii) Using a dummy for “application resulting in continuation-in-part” as an alternative
measure of complexity (as a sign of the firm continuing to work on that invention), and a
dummy for “resulting in divisional” as an alternative measure of work quality (since this action
is usually the result of a restriction requirement) revealed that these were more likely to occur
for applications filed at month-ends (Table B22).
(ix) Including the daily volume of filings (for each firm) as an additional control did not change
the substance of the results (Table B23).
(x) Using alternative measures of size (assets, sales, and employment) did not affect the results
relating to the firm size–clustering association. When jointly included with patent volume (our
measure of firm size), only patent volume remained significant (Table B24).
(xi) The firm size–deadline clustering association is robust to using application data (Table
B25A), and the inclusion of technology class or inventor fixed effects (Table B25B).
(xii) We verified robustness of the quality results to splitting the sample by more and less than
20 claims (Tables B26A–B). Interacting the month-end dummy with a dummy for more than 20
claims showed positive interaction terms, though they were generally insignificant (Tables
B26C).
(xiii) The quality results were robust to including joint firm–claims fixed effects (Table B27A).
(xiv) The complexity and quality results were robust to including joint firm–monthly volume
for the firm fixed effects (Table B27B), and to using firm–year–month fixed effects instead of
firm fixed effects (Table B28).
30
IV. Robustness Checks
TABLE B1: Key Results
Table in Text D1 D3 D5 N
Work clustering
1 (7-9) Mean daily share
0.389**
(0.108)
0.353**
(0.058)
0.336**
(0.049)
366
4 (1-3) Log firm-year
size
2.842*
(0.582)
5.533*
(0.702)
6.542*
(0.676)
2,878,229
Work sorting at month-ends
2 Log claims 17.354*
(3.156)
9.434*
(2.259)
7.953*
(2.155)
1,801,602
2 Log cites (5 yrs.) 15.018*
(3.018)
6.116*
(1.968)
3.897+
(1.682)
2,633,488
2 Renewal (3.5 yrs.) 2.823+
(1.253)
2.051+
(0.859)
2.161*
(0.765)
1,938,296
Work quality at month-ends
Work process-based measures
3 (2-5) App. incomplete 45.820**
(10.737)
36.129**
(7.063)
29.983**
(5.550)
92, 533
3 (2-5) Sep. inv. oaths 48.661**
(8.605)
37.288**
(6.208)
32.278**
(4.871)
92,533
3 (2-5) Add. filing fees 46.607**
(9.403)
36.410**
(6.745)
30.267**
(5.405)
92,533
Work outcome-based measures
3 (2-5) Exam. add. cites 0.254
(3.388)
4.140+
(1.980)
4.095*
(1.788)
627,209
3 (2-5) App. approved -1.441
(2.239)
-4.578*
(1.497)
-4.456*
(1.153)
785,051
3 (2-5) Log review time 23.139**
(4.015)
14.246**
(2.526)
9.011**
(2.403)
557,383
This table presents some of the main results (“Key” results henceforth) in one place for convenient comparison with other
results presented in appendix tables. Refer to the main text for details on these specifications.
31
IV. Robustness Checks
TABLE B2: Construction of the samples (Tables 1-4 in the main text)