consistent with the invention: Evidence from Japan

Transcription

1 Contribution of patent examination to making the patent scope consistent with the invention: Evidence from Japan March 2016 Yoshimi Okada, Yusuke Naito Institute of Innovation Research, Hitotsubashi University Sadao Nagaoka Tokyo Keizai University Abstract An important task of the patent office is to narrow the overbroad claims so that they become consistent with the invention. This study empirically analyzes how significantly the patent office delivers this important function through its patent examination. We found that the patent right scope very often (i.e. two thirds of the granted patents) gets narrower as the outcome of patent examination process and becomes more consistent with the invention. Both the incidence and the extent of such narrowing increase when the applicant chooses broader claims, and decrease when the quality of prior art disclosure by applicant is higher. Key words: Patent scope; Examination; Claim length; Disclosure JEL: O34, O31, O38 p. 1

2 1. Introduction Delineating the patent scope consistent with the contribution of the disclosed invention to the state of the art is one of the crucial requirements for the patent systems in promoting innovation. An overbroad claim restricts the use of knowledge which is already in the public domain or preempts the knowledge domain not yet invented, while an over-narrow claim weakens incentive for an invention with a large inventive step. It is very important to note that a firm has an incentive to make broad patent claims, although overbroad claim can cause a delay of a patent grant, which can serve as a disincentive for such behavior. Given the existence of such incentive of a firm, an important task of the patent office is to narrow the overbroad claims so that they become consistent with the invention. That is, the examiner s role in the patent examination is not just to assess the patentability of the invention but also, perhaps more importantly, to guide the applicant so that they properly narrow down such broader claim so as to make the claim commensurate with the contribution of the invention, i.e. its contribution to the state of the art. This study empirically analyzes how significantly the patent office serves this important function through its patent examination. To the best of our knowledge, there are no full-scale econometric analysis of this issue. This is significantly because most economic analyses of the patent system makes a simple assumption that a firm files a patent application with the claims consistent with the invention. Thus, they assume that the task of a patent examiner is simply to assess the novelty, the inventive step and utility of the invention and to conclude whether the invention is patentable or not. However, this view of a patent examination misses a crucial function of the patent examination, which is to lead applicants to properly delineate the patent scope consistent with the invention. However, there are a limited number of prior studies investigate examination quality, using patent length as an indicator of the patent scope. Malackowski & Barney (2008) used the average number of words per independent claim in an issued US patent as a proxy of the quality of patent examination by the USPTO. They mentioned that the number of words in independent claims increased through examination process from on filing to on issuance around the time of their investigation. They also found that independent claims of the US patents issued in 2007 had 4.4 percent more words compared to that of patents issued in 2003, which they say is one of the evidences that the quality of patent examination had not decreased during that period. While they used the claim length as a proxy for quality of patent examination procedure, they did not identify the factors causing the increase of claim words after p. 2

3 patent examination. Webster et al.(2014) investigated incidence of change in independent claims in 236 EP patents and 82 JP patents. However the purpose of their investigation is investigating the violation of respecting the national treatment principle that is stipulated in the TRIPS agreement. Osenga (2011) focused on the number of words in claims. But he used the claim length as a proxy of readability or comprehension of claims. This study investigates how the patent office narrows the scope of the patent right. We use the claim length (more exactly the inverse of the claim length) of the first claim (claim No. 1) as a measure of patent scope. We then examine how the difference between the scope of the claim of the granted patent relative to that of the initial claim at the time of filing the patent application are determined by the broadness of the initial claims at the time of filing as measured by the claim length and the number of claims, and the quality of the disclosure of relevant prior arts. Briefly we found that the patent right scope very often (i.e. two thirds of the granted patents) gets narrower as the outcome of patent examination process and that an applicant is more likely to get a patent in its initial claim or less likely to need to significantly amend the claim, if the initial claim is narrower and if the prior art is well disclosed in the patent application. This paper is organized as follow. Section 2 presents a short description of the legal background. Section 3 offers an explanation of the comprehensive database of the Japanese patent applications and grants newly developed for this research project and the basic statistics. Section 4 presents hypotheses and estimation models, which assess how the incidence of the narrowing of the claim as well as the degree of such narrowing can be explained by the applicant backward citations disclosed in the patent application documents as well as the broadness of initial claims. Section 5 presents estimation results and discussions. Section 6 concludes, with discussions of policy implications. 2. Legal background The scope of patent rights are described in the documents called "claims." Each words in a patent claim has a role to limit the scope of the patent right. Briefly, the process of patent examination is as follows. The patent examiner presents prior arts that denies the novelty or inventive step of the focal invention to the applicant and the applicant amends the claims to narrow down the scope of the exclusive right so as not to cover the prior arts to obtain a patent. There are two ways to narrow down the patent scope. One is to add new p. 3

4 elements to restrict the scope of the concepts of the claims. This is the most frequently used way in inventions of appliances. The scope of the invention is presented by the logical multiplication of the scope of each items. We call this type of claim structure cascaded type. The other one is to eliminate some of the items which are listed in the claims as the choices that a person will need to make when using the invention. The second way is frequently used in inventions of materials or chemical compositions, especially in inventions described by Markush type claims. The scope of the invention is presented by the logical addition of the scope of each items. We call this type of claim structure parallel type Usually and conventionally the first claim (claim No. 1) conveys the broadest inventive concept. We therefore focus on the first claim. Because the Japanese language does not pose spaces between words, it is hard to automatically count words in Japanese sentences. We use the number of characters instead of the number of words for the metric of claim length. Hereinafter, we define the claim length as the number of characters of the first claim of a patent application or the grant. In general, because the number of elements to limit the scope of a concept increases with the increasing number of characters in a claim, one can say that there is a negative correlation between the size of the number of characters in a claim and the broadness of the conceptual scope of the claims in cascaded type claims. If we focus only on the inventions of cascaded type claims, we will have a strong correlation between the scope of patent and the claim length. According to this idea, we measure the broadness of the scope of claim by the inverse of claim length. 3. Data construction and descriptive statistics (1) Data constructed The base data called PATBIB were prepared from the patent databases maintained by Artificial Life Laboratory, Inc. The data were basically prepared from both the data called td-5, which composed of text data of pre-grant patent application publications and the granted patent publications, and the data called Pat-R which composed of bibliographic data and examination process data that were originally provided by the Japan Patent Office The PATBIB contains the data of the number of characters in the first claim as well as the bibliographic data and examination data. It is likely that the length of the claim measures the patent scope in a different manner between product patent and process patent. This study focuses on product patent, which accounts for more than 80% of the whole patents (to be explained in the following subsection). In order to divide patents into the two categories of product p. 4

5 patents and "process patents," we first prepared learning data for our computer program and made it to automatically divide the category of inventions of first claims based on the learning data. After adjusting the program, we randomly chose 460 sample patents and verified there were no errors in category data of the first claims. The applicant citations are the patent documents that were cited and disclosed by the applicants in the specification of the patent applications. The applicant citation data were included in the above td-5 database. In order to address the variations in notations of applicants names, we utilized NISTEP dictionary of Japanese company names and the connection table to IIP patent database, both of which were provided by the National Institute of Science and Technology Policy, the Ministry of Education, Culture, Sports, Science and Technology (NISTEP). Each of major applicants has its unique NISTEP id. Utilizing the id data in NISTEP dictionary, we can identify the same company even if there were changes in the names of companies or M & A. We used PATSTAT(2014 Autumn version) that were provided by the European Patent Office in order to obtain the information of the existence of patent families, namely the corresponding US patent applications that were filed to the USPTO on the same inventions. We used the Japanese patent applications only to those whose effective filing date is between January 1993 and August Because there was a law change on and after September 2002 on disclosure, we chose the last month of our sample in order to eliminate the influence of this law change taking into account of the maximum 12 months priority. The first month of our sample is chosen only for the availability of text data. We also eliminate the divisional applications from our data sets to make things simple. We divide technological fields according to the definition by IIP patent database (see appendix for the concordance table to IPC). It divided the technological fields into 33 technological fields. Then the technological fields are grouped to six large technology fields namely, Chemical, Computers & Communications, Drugs and Medical, Electrical & Electronic, Mechanical, and Others. We focus on the four sectors, namely Computers & Communications, Electrical & Electronic, Mechanical and Others where cascaded type claim is expected to be dominant and the amount of parallel type claims is relatively small. We also eliminate the patents and the patent applications that involves chemical formula, mathematical formula or tables in the first claim, utilizing the tag information in patent database in order to collect only patents with cascaded type claims as much as possible. We also prepared the learning data of first claim that contains chemical formula, mathematical p. 5

6 formula or tables in the first claim but that have no tag information. Based on these works, we constructed the data with hopefully the patent applications and grants with only cascaded type claims. (2) Distribution of claim length of applied and granted patents Table 1A shows the descriptive statistics of the claim length on granted patents in each technological fields. Product inventions account for more than 80% of the inventions in each of the four fields. Computers & communication have the patent grants with the longest average claim length. The mean value of claim length is larger for product inventions compared to that of process inventions in all major fields and the degree of the difference in the mean value of the claim length between the product inventions and the process inventions differs among technological fields. The composition ratio between product inventions and process inventions also differs among technological fields. Therefore, we will focus on only the product inventions in the rest of our paper. Figure 1 shows the distribution of the natural logarithm of the inverse of claim length of granted patents for product inventions in all fields (aggregation of product patents in Computers & Communications field, Electrical & Electronic field, Mechanical field and Others). The shape of the distributions look close to that of normal distribution. Figure 2 shows the comparison of the distribution of the natural logarithm of the inverse of the claim length between before and after examination for product inventions in both Computers & Communications and Electrical & Electronic sectors. The blue plots are for initial claim length at filing; the red plots are for initial claim length at filing for applications that were eventually granted; and the green plots are for claim length of granted patents. It is clear that all the shapes of the distributions look close to that of normal distribution and the distributions significantly moves to the left due to examinations which means the average scope of the granted claim became narrower compared to initial claim after examination. (3) Examination outcomes by the initial claim length Table 2A and 2B show the outcome of patent examinations for the four sectors. Not granted" includes rejected, withdrawn or abandoned, and dismissed 1. Granted cases are divided into three categories according the change of the scope of claim before and after the examinations, namely, broadened, unchanged and narrowed, corresponding to claim-length-decreased case, claim-length-unchanged case and 1 Even if the applicant received the decision to grant a patent, the application is dismissed if the applicant does not pay the fee for registration. p. 6

7 claim-length-increased case respectively. The cases where the claim length became shorter and the scope became broader is very small( around 2 to 3 %) of the total patent applications that were requested for examinations, suggesting that we can effectively succeeded in eliminating the patents with parallel type claims from our sample. We drop those patent applications which experienced the decrease of the claim lengths from our sample for our following estimations. In large majority of cases, the claim length became longer after the patent examinations: 67% of the total eventually granted patents in these 4 major technology sectors. This ratio varies only a little across four sectors. Thus, it is very clear that patent examination plays a very important role in delimiting the scope of the patent in light of the contribution of the invention in all these technology fields. Figures 3A and 3B show how these examination outcomes differ by the length of initial claim for Computers & Communications and Electrical & Electronic sectors. The figures show that the grant probability decreases and the narrowing probability increases significantly with the scope of the claim (inverse of claim length) at patent application. It indicates that the patent examiner is more likely to narrow the claim when the applicant pursues a broad claim (a claim with short descriptions), as we expect. 4. Hypotheses and estimation models 4.1 Hypotheses An examiner assesses whether the claim of the patent is consistent with the net contribution of the invention disclosed. If the claim is too broad relative to the net contribution of the invention to the state of the art, the examiner will issue an initial rejection which forces the applicant to amend the claim so as not to cover what is in the public domain. While we do not directly observe the threshold patent scope allowed for the invention, the applicant is more likely to face an initial rejection when it asks for broad claims in terms of the scope of each claim as well as the number of claims. Thus, we have the following hypothesis 1, which we can test. Hypothesis 1 Broad claims and examination A patent application with broader claims (with a short claim description and a more number of claims) is more likely to experience the narrowing event. In addition, given the amendment, the extent of narrowing is larger when the claims are broader. Furthermore, an application with more disclosures of relevant prior art (that is, p. 7

8 higher quality disclosure) is more likely to have the claims consistent with the contribution of the disclosed invention to the prior art. Thus, we have the following testable hypothesis 2 on the effect of the quality of the disclosure of relevant prior art on examination. Hypothesis 2 Applicant disclosure of relevant prior art and examination A patent application with higher quality disclosure of prior art experiences less the narrowing event. In addition, given the amendment, the extent of narrowing is smaller for an application with higher quality disclosure of prior art. In testing hypothesis 2, we need to develop a measure of the disclosure quality. Our main measure is the share of the co-cited patents (prior patents cited both by the applicant and by the examiner) in the entire patent citations made by an examiner examining the focal patent (q) (in the appendix II, we present the results using the number of applicant citations as a measure of disclosure quality as a robustness check). Figure 4 shows the citation categories, where a is the number of prior art cited only by the applicant, b is the number of prior art cited both by the applicant and by the examiner, and c is the number of prior art cited only by the examiner. If we define q=b/(b+c), q can be interpreted as the quality of prior art disclosed by the applicant, measured by the coverage of examiner citations. That is, if the applicant anticipates all examiner citations and discloses all of them, the disclosure quality is 1 (100%). On the other hand, if the applicant anticipates none of them, it is zero even if it refers to many prior patent literatures. We have developed this measure by matching the examiner and the applicant citations. 4.2 Estimation issues and strategy Our new measure of the disclosure quality brings us two estimation issues. First, it is likely to be endogenous, given that it is also affected by the patent examiner. If an examiner happens to fail in finding some relevant prior art, variable q is high not because of high disclosure quality of the applicant but due to the mistake by the examiner. Simultaneously, such application is less likely to experience the narrowing event, so that we will observe a negative correlation between variable q and the narrowing event. This is a spurious correlation. If such cases are frequent, the OLS regression over-estimates the effect of disclosure quality and the coefficient of q is biased upward. Second, this variable q is a very imperfect measure of disclosure quality, since p. 8

9 it is based only on the share of the co-cited patent literatures. In particular, it totally neglects the contents of the relevant prior art, thus the variations of prior art in terms of its importance in determining the contribution of the invention. For an example, when the examiner makes a decision to grant, the examiner is not strictly instructed by the examination guidelines as to what to cite for the reference. So there occur cases that a patent examiner does not cite the applicant citations even if they are as relevant as what examiner actually cited and vice versa. If such measurement errors are significant, the OLS regression underestimates the causal effects of the quality of applicant prior art disclosure. We will use instrumental variables to address these two problems. The instruments are the past (one year average) disclosure quality in terms of q as well as the past (one year average) number of applicant citations which would be explained in more detail below. Given that each patent examination is independent from the past patent examinations (note that we dropped divisional applications from our sample) and that an examiner is essentially randomly assigned for a new patent application in a given technical field, the past (one year average) disclosure quality in terms of q should be uncorrelated with the examination performance of the current patent application, except that there is a persistence of the disclosure and claiming strategy of a firm. Because we use the average data, we eliminated the patents of the applicants who obtained less than 400 patents in the focal 4 major technological sectors. 4.3 Estimation model The estimation model for the narrowing event is as follows: = + _ h( ) + _ _ ( ) + _ _ + _ _ _ + _ _ + _ _ + The first dependent variable is the dummy for narrowing event, which is set to 1 if the claim length became longer after the examination and is set to 0 if the claim length did not change by the examination. Among the explanatory variables, variable q is the coverage of the examiner citations by the applicant citations. As an instrument, we use the past one year average of q and variable ln_ave_n_backward_citation which stands for the natural logarithm of the past one year average of the number of applicants backward citation. We utilize the following two variables to measure the broadness of the patent scope claimed:. variable ln_inverse_claim_length(intitial) and variable ln_n_claims(initial). Variable ln_inverse_claim_length(intitial) stands for the natural logarithm of the inverse of the initial (that is, at the time of patent p. 9

10 application to the JPO) claim length (the first independent claim if there are two or more claims). Variable ln_n_claims(initial) stands for the natural logarithm of the initial number of claims. We utilize the following variables to measure the technological or economic importance of the inventions: Variable ln_n_inventor stands for the logarithm of the number of inventors; Variable ln_n_forward_citation stands for the natural logarithm of the number of applicants forward citations; The variable subseq_appln_flg is set to one when the subsequent divisional applications for the focal patent exists and set to 0 otherwise; The variable US_Grant_flg is set to 1 if there is at least a corresponding US patent and set to 0 otherwise. Descriptive statistics for the data are shown in table 3. The second dependent variable is a continuous variable for the Extent_of_narrowing that indicates the magnitude of change in claim length due to the examination, which we define by the following formula: Extent_of_narrowing = ln(claim length at grant initial claim length +1). If we assume that the length of a paragraph required for describing one additional restriction on the scope of the patent is fixed, the absolute difference between the claim length at the grant and the initial claim length, rather than their ratio, indicates the number of restrictions introduced due to the examination. We adopt this interpretation. The estimation model for the extent of narrowing of the claim uses the same explanatory variables, except for the length of claim, since the dependent variable Extent of narrowing directly uses this variable 2. The sample for the estimations are the patent grants which experienced amendments. 5 Estimation results and discussions In the regression analysis we only consider those applications the applicants of which are identified by the NISTEP applicant id. And we removed co-owned patents in order to use fixed effect of applicant model. We will present OLS and IV results for the incidence of narrowing event, then those for the extent_of_narrowing and robustness check. There are two IV models. The first model uses both of the two instruments: the average of past q that is calculated for the period of preceding one to twelve months and the logarithm of the average of the number of the applicant s backward citations that is calculated for the period of preceding one to twelve months, each for all combinations of an applicant and a technology sector pair. The second model uses only 2 Thus, for whatever a reason, the length of claim at the application (for an example, the use of the non-abbreviated words) tends to generate a negative correlation between the claim length at the application and the measure for the extent of narrowing. p. 10

11 the first instrument. 5.1 Incidence of narrowing event First, Table 4 shows the results of the OLS and the IV estimations of the incidence of narrowing against q, two claim variables at application and the control variables including the applicant dummies (applicant fixed effects). The estimation models are based on the data pooling four technology sectors. The descriptive statistics is given in Table 3. The results of the regression of the first stage (Model (3) and (5)) indicate that the instrumental variables explain significantly the variation of variable q, showing that the disclosure policy of an applicant has significant persistence over time. (Table 3 and Table 4) The coefficient of ln_inverse_claim_length and that of the ln_n_claims are positive and highly significant for all of the three (OLS and the two IV second stage) estimations. The estimated coefficients are similar among these thee estimations. Thus, it is clear that the patent application with broad claims is significantly more likely to experience a narrowing decision by the examiner, supporting hypothesis 1. The elasticity with respect to the claim length is large (around 0.2): one standard deviation reduction of the number of characters of the first claim (around 0.5 in logarithmic term) will result in around 8 to 10 % points increase of the narrowing event. As for the number of claims, one standard deviation increase (0.8 in logarithmic term) will results in 3 to 4% points increase of the narrowing event. The coefficient for disclosure quality variable q is negative and highly statistically significant for all models, suggesting that when the applicants cited more number of relevant prior patent literature which the patent examiner will also cite, it is more likely that the claim length remain unchanged. Higher quality disclosure, which is likely to be accompanied with claim construction consistent with relevant prior art, reduces the probability of the event for the patent office to narrow the claim, supporting hypothesis 2. The size of the estimated coefficient is much larger for IV models (more than 5 times larger): for the OLS estimation (Model (1)) vs and for IV estimations (Model (2) and (4) respectively). One standard deviation increase of the disclosure quality (0.16) results in 12% points increase of the narrowing event, according to Model (3) (IV), showing that the effect of the applicant disclosure on the incidence of the narrowing event is large. The significant underestimation by the OLS also indicates that there exists a serious measurement error of quality which tends to overwhelm endogeneity bias for estimating q. The forward citation has a positive and highly significant coefficient, with the p. 11

12 coefficient around The number of inventors also has a positive and highly significant coefficient around There are no significant differences between the OLS and the two IV results for estimating these coefficients. These results imply that a patent application with more technological contents is more likely to face initial rejections for amendments, controlling for the other variables. This is not surprising, considering that a technologically important invention is accompanied with broader claims in terms of contents, controlling for the number of claims and for the length of the first claim, so that an examiner is more likely to find inconsistencies between the invention and the claims. The coefficients of the variable subseq_appln_flg is positive and statistically highly significant. The value is quite high and close to 0.1 for IV models. This finding means that a patent with a future divisional application is more than 10% likely to experience the narrowing event. Under the Japanese patent law system, an applicant was not allowed to file a divisional application based on the focal patent after he/she received the decision to grant a patent before April 2007 when the patent law was revised to allow it. Therefore if an applicant wants to reserve the opportunity of filing a divisional application based on the focal patent, he/she make the initial claim broader 3, so that the first action made by a patent examiner is notification of reasons of refusal. Table 5 shows the summary results of instrumental variable regression for the data of each of four technology sectors. The descriptive statistics is given in Apendix Table II-1. According to Table 5, both the coefficient of ln_inverse_claim_length and that of the ln_n_claims are positive and highly significant in all technology sectors. The estimated coefficients are broadly similar across sectors. In addition, the estimated coefficient for q is also negative and significant in all technology sectors, although it is highly significant only in Electrical & Electronic field. These results support both hypothesis 1 and 2.The weaker significance of the coefficient of q is likely to reflect the fact that the instrumental variables are based on firms and not on patents, so that the effective sample size is much smaller as for estimation of the coefficient of q. (Table 5) 5.2 Extent of narrowing Table 7 shows the summary of the results of OLS and IV regressions of the extent of narrowing against q, the number of claims and other controlling variables with fixed 3 Also, it sometimes happen that the applicant get an initial patent with a narrower claim and acquires a broader claim by a divisional application that are filed based on the focal patent application, because sometimes they want to get some patents in early stage at any hand for inventions they think important. p. 12

13 effects of applicant dummies, based on the pooled data. The descriptive statistics is given in Table 6. The coefficient of the ln_n_claims is positive and highly significant for OLS and the two IV second stage estimations. The estimated coefficients are almost same among the three estimations. Thus, it is clear that the applicant which seek more number of claims are more likely to experience a larger extent of narrowing of the first claim, supporting hypothesis 1. The elasticity with respect to the claim length is large (around 0.13): one standard deviation reduction of the number of claims (around 0.79 in logarithmic term) will result in around 10 % points increase of the number of characters that are added by the amendment made in examination procedures. The coefficient for variable q is negative and highly statistically significant for all three estimations, showing that when the quality of the applicant s disclosure is better and therefore the coverage of examiner citations by applicant citations is high, the extent of narrowing also becomes smaller, supporting hypothesis 2. The applicants that searched relevant prior art well before patent applications and designs the claim consistently with prior art are more likely to acquire the scope of the patent protection which is closer to what he requested at the filling. Similar to the estimation results for the incidence of narrowing, IV estimation gives a significantly larger coefficient to the effect of disclosure quality q (more than seven times larger), suggesting the importance of controlling for the measurement error of disclosure quality. As for control variables, the coefficient for the number of inventors is negative (around -0.02), unlike that for the incidence of narrowing, and highly significant, and there are no significant differences between the OLS and the two IV results. The different signs of the coefficient for the number of inventors for the narrowing event and for the extent of narrowing implies that an invention with significant technological content measured by the number of inventors experiences less restrictions of the claims (that is, the applicant needs to add a smaller number of restrictions to the claim), even though a narrowing event itself is more likely to happen, when we control for the number of claims and its scope. The coefficient for the forward citation became insignificant for the extent of narrowing, unlike for the incidence of narrowing. There are no significant differences between the three estimations. (Table 6 and Table 7) Table 8 shows the summary results of instrumental variable regression for the data of each of four technology sectors. The descriptive statistics is given in Apendix Table II-2. According to Table 8, the coefficient of the ln_n_claims is positive and highly significant in all technology sectors. The estimated coefficients are broadly similar across sectors. In addition, the estimated coefficient for q is also negative and p. 13

14 highly significant in all technology sectors except for in Mechanical field where the estimated coefficient for q is marginally significant. These results support both hypothesis 1 and 2. The weaker significance of the coefficient of q is again likely to reflect the fact that the instrumental variables are based on firms and not on patents, so that the effective sample size is much smaller as for estimation of the coefficient of q. 5.3 Robustness check As a robustness check, we did estimations, using the number of the prior patents cited by the applicant as a measure of disclosure, replacing the disclosure quality variable q, which depends partly on the examiners. As the applicant increases the number of such prior art citations, we expect that the disclosure quality also improves. The results for the incidence of narrowing are shown in Appendix table III-1 (descriptive statistics) and III-2 (estimation results). The qualitative results are the same as those presented in Table 4. According to the estimation results, the patent application with more disclosures is less likely to experience the restriction of the scope of the first claim and that with broader claims (the scope of the initial claim and the number of claims) is more likely to experience such restriction. The results for the extent of narrowing are shown in Appendix table III-3 (descriptive statistics) and III-4 (estimation results). The qualitative results are the same as those presented in Table 7. According to the estimation results, applicants that disclose more prior art before patent applications are more likely to acquire the scope of the patent protection which is closer to what he requested at the filling. Applicants which seek broader claims,i.e. more number of claims, are more likely to experience a larger extent of narrowing of the first claim. 6. Conclusion Delineating the patent scope consistent with the contribution of the disclosed invention to the prior art is one of the crucial requirements for the patent system in effectively promoting innovation. Given the incentive of a firm to claim the scope of the patent right as broad as possible, an important task of the patent office is to narrow the overbroad claims so that they become consistent with the invention. This study empirically analyzed how significantly the patent office delivers this important function through its patent examination. The existing economic literature has not conducted a full-scale econometric analysis on this issue, to the best of our knowledge We used the change of the length of the first claims before and after the examination in order to investigate how the broadness of the initial claims at the time of filing and p. 14

15 the quality of the disclosure of relevant prior arts drives the change of the length of the first claim. We focused on product inventions in 4 major technology areas where cascaded type claim is dominant. We found that the claim length very often (i.e. two thirds of the granted patents) becomes longer as a result of patent examination. Such adjustment of the claim is significantly less likely to happen when the applicant discloses more number of relevant prior arts, controlling for the endogeneity of disclosure quality. It is more likely to happen when the applicant chooses broader claims (that is, a short claim and more number of claims). The same set of variables also affects the magnitude of the change in the claim length due to the examination. Thus, we found that there is a major contribution of patent examinations to making the claims consistent with the inventions. The results of this study show that the patent office plays a very significant role in delineating the scope of a patent, which is a cornerstone of the patent system in promoting innovation. This reaffirms the importance of the quality of the patent examination and the supporting infrastructure for the patent office. The study results also suggest that policies supporting the applicants to improve the quality of prior art disclosure will matter for patent examination efficiency (for an example, less frequent amendments). Thus, it is also important for the patent office to provide prior art information to the applicants at low cost and through Internet by improving the infrastructure of patent information search system. Acknowledgment This work was supported by JSPS KAKENHI Grant Number We would like to thank JSPS for granting this aid. We would like to thank research assistant Junko Morikawa for her help in the sample check investigation on category data of the first claims. p. 15

16 Major References Abiko Gen & Yoshitoshi tanaka (2010), The Study of Variance of Decision Making to Claim Influence on Usefulness, Journal of Intellectual Property Association of Japan Vol.7 No.1 p.54-67( ) Abiko Gen, Kenichiro Ayaki, Toshimitsu Kataoka, Junichi Akama, Takashi Nakai, Shiro Ataka (2012), Quantitative Analysis of the Size of Technical Scope on the Patent of LED, Journal of Intellectual Property Association of Japan Vol.8 No.3 p ( ) Goto Akira and Kazuyuki Motohashi Construction of a Japanese Patent Database and a first look at Japanese patenting activities, Research Policy, Volume 36, Issue 9, November 2007, Pages , Hall, B. H., Jaffe, A. B., & Trajtenberg, M. (2001). The NBER patent citation data file: Lessons, insights and methodological tools (No. w8498). National Bureau of Economic Research. Malackowski, J., & Barney, J. (2008, June). What is Patent Quality? A Merchant Banc's Perspective. Les Nouvelles, what-is-patent-quality-a-merchant-banc-s-perspective NISTEP, Company name dictionary URL: information-infrastructure NISTEP, Connection table to other databases URL: Osenga, Kristen. "Shape of Things to Come: What We Can Learn from Patent Claim Length." Santa Clara Computer & High Tech. LJ 28 (2011): 617. Webster E., Jensen P. H., and Palangkaraya A., Patent examination outcomes and the national treatment principle, RAND Journal of Economics Vol. 45, No. 2, Summer 2014, pp p. 16

17 Figure 1 Distribution of -ln(claim length) of granted patents for product inventions All fields (product inventions) frequency ln(claim length) Figure 2 Comparison of the distribution of the natural logarithm of the inverse of claim length between before and after examination C & C (product inventions) 9.0% 8.0% 7.0% 6.0% 5.0% 4.0% 3.0% 2.0% 1.0% 0.0% narrow -ln(claim length) Initial claim for both granted and not granted Initial claim for eventually granted broad 9.0% 8.0% 7.0% 6.0% 5.0% 4.0% 3.0% 2.0% 1.0% 0.0% narrow E & E (product inventions) -ln(claim length) broad Initial claim for both granted and not granted Initial claim for eventually granted Granted claim Granted claim p. 17

18 Figure 3 Distribution of ln(claim length) and outcome of the examination C & C 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% narrow -ln(claim length(initial)) broad broadened unchanged narrowed Not granted E & E 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% narrow -ln(claim length(initial)) broad broadened unchanged narrowed Not granted p. 18

19 Figure 4 Degree of quality of the applicant disclosed prior art p. 19

20 Table 1 Claim length (number of characters in claim 1) of granted patents technology fields category of invention n mean sd p1 med p99 Computers & Communications product 124, process 21, Electrical & Electronic product 152, process 39, Mechanical product 138, process 28, Others product 97, process 14, p. 20

21 Table 2A Composition of the outcome of patent examination Granted Technological field broadened unchanged narrowed Not Granted Total Computers & Communications 6,490 31,304 86, , ,185 Electrical & Electronic 7,068 40, , , ,298 Mechanical 4,626 38,577 97,656 95, ,623 Others 3,127 29,368 66,226 73, ,351 Total 21, , , , ,457 "Not granted" includes rejected, withdrawn abandoned, and dismissed Table 2B Composition ratio of the outcome of patent examination Granted Technological field broadened unchanged narrowed Not Granted Total Computers & Communications 2.6% 12.7% 34.9% 49.8% 100% Electrical & Electronic 2.5% 14.0% 38.9% 44.6% 100% Mechanical 2.0% 16.3% 41.3% 40.5% 100% Others 1.8% 17.0% 38.4% 42.7% 100% Total 2.3% 14.8% 38.4% 44.6% 100% "Not granted" includes rejected, withdrawn abandoned, and dismissed p. 21

22 Table 3 Descriptive Statistics for incidence of narrowing reported in Table 4 based on pooled data p. 22

23 Table 4 Summary results of OLS and instrumental variable regression for the incidence and extent of narrowing based on the pooled data Dependent Variable: incidence of narrowing (narrowed=1, unchanged=0) (1) (2) (3) (4) (5) Pooled Pooled Pooled Pooled Pooled OLS IV IV IV IV second stage first stage second stage first stage Independent variables q = b/(b+c) -.134*** -.733*** -.919*** (.0049) (.132) (.153) av_q.217***.285*** (.0170) (.0156) ln_av_n_backward_citation.0200*** (.00198) ln_inverse_clim_length(initial).187***.177*** ***.174*** *** (.00155) (.00273) ( ) (.00304) ( ) ln_n_claims(initial).0469***.0480***.00181***.0484***.00185*** (.00118) (.00123) ( ) (.00126) ( ) ln_n_inventor.00365***.00595***.00372***.00667***.0038*** (.00138) (.00151) ( ) (.00156) ( ) ln_n_forward_citation.0179***.0182*** *** (.00105) (.00108) ( ) (.0011) ( ) subappln_flg.105***.101*** ***.0998*** *** (.00361) (.00379) (.00135) (.00389) (.00135) US_Grant_flg.00576***.00890***.00507***.00987***.00519*** (.00189) (.00205) ( ) (.00212) ( ) effective filing year yes yes yes yes yes 33_clasiffication yes yes yes yes yes effective filing year * 33_clasiffication yes yes yes yes yes applicant FE FE FE FE FE Observations 299, , , , ,934 R-squared within Number of patent owners Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.10 p. 23

24 Table 5 Summary results of instrumental variable regression for the incidence of narrowing Independent variables Dependent variable: incidence of narrowing (narrowed=1, unchanged=0) (1) (2) (3) (4) (5) (6) (7) (8) C & C C & C E & E E & E mech. Mech. Others Others IV IV IV IV IV IV IV IV second stage first stage second stage first stage second stage first stage second stage first stage q = b/(b+c) -.612* -1.59*** -.820** -.705* (.324) (.379) (.402) (.42) av_q.139***.179*** *** (.0429) (.0359) (.0361) (.0366) ln_av_n_backward_citation.0351***.0185***.0365***.0122** (.00604) (.00436) (.0079) (.00615) ln_inverse_clim_length(initial).201*** ***.142*** ***.172*** ***.194*** *** (.0048) ( ) (.00657) ( ) (.00982) (.00139) (.00924) (.00175) ln_n_claims(initial).0432***.0017**.0482***.00245***.0548*** ***.00238* (.00218) ( ) (.00246) ( ) (.00265) (.00107) (.00371) (.00139) ln_n_inventor ***.00961***.00476*** ***.00884** (.00285) ( ) (.00324) ( ) (.00349) (.00123) (.00383) (.00145) ln_n_forward_citation.0218*** ***.00164***.0129*** *.0217*** (.00206) ( ) (.00204) (.00061) (.00239) ( ) (.00313) (.00121) subappln_flg.0985*** **.0913*** ***.114*** *** (.0061) (.0019) (.00797) (.0022) (.0094) (.00385) (.0109) (.00418) US_Grant_flg ***.0210***.00630***.00871*.00515*** *** (.00334) (.00104) (.00426) (.0011) (.00476) (.00175) (.0086) (.00304) effective filing year yes yes yes yes yes yes yes yes 33_clasiffication yes yes yes yes yes yes yes yes applicant FE FE FE FE FE FE FE FE Observations 89,254 89,254 97,365 97,365 73,107 73,107 40,208 40,208 R-squared Number of patent owners Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.10 p. 24

25 Table 6 Descriptive Statistics for extent of narrowing reported in Table 8 based on pooled data N mean sd p99 p1 Extent of narrowing q = b/(b+c) ln_n_claims(initial) ln_n_inventor ln_n_forward_citation subseq_appln_flg US_Grant_flg av_q ln_av_n_backward_citation p. 25

26 Table 7 Summary results of OLS and instrumental variable regression for the extent of narrowing based on the pooled data Independent variables Dependent variable: extent of narrowing (= ln(claim length(grant) - claim_length(initial) + 1)) (1) (2) (3) (4) (5) Pooled Pooled Pooled Pooled Pooled OLS IV IV IV IV second stage first stage second stage first stage q = b/(b+c) -.213*** -1.53*** -1.53*** (.0153) (.412) (.471) av_q.203***.267*** (.0186) (.0171) ln_av_n_backward_citation.0182*** (.00209) ln_n_claims(initial).128***.129***.00102**.129***.00103** (.00342) (.00349) ( ) (.0035) ( ) ln_n_inventor *** ***.00296*** ***.00304*** (.00396) (.00422) ( ) (.00428) ( ) ln_n_forward_citation (.00299) (.00305) ( ) (.00305) ( ) subappln_flg.0987***.0920*** ***.0920*** *** (.00947) (.00985) (.00131) (.00992) (.00131) US_Grant_flg *** *** (.00538) (.00606) ( ) (.00623) ( ) effective filing year yes yes yes yes yes 33_clasiffication yes yes yes yes yes effective filing year * 33_clasiffication yes yes yes yes yes applicant FE FE FE FE FE Observations 224, , , , ,581 R-squared Number of patent owners Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.10 p. 26

27 Table 8 Summary results of instrumental variable regression for the extent of narrowing Dependent variable: extent of narrowing (= ln(claim length(grant) - claim_length(initial) + 1)) (1) (2) (3) (4) (5) (6) (7) (8) C & C C & C E & E E & E mech. Mech. Others Others IV IV IV IV IV IV IV IV second stage first stage second stage first stage second stage first stage second stage first stage Independent variables q = b/(b+c) -3.06** -2.89*** -2.47* -4.24*** (1.42) (.942) (1.4) (1.47) av_q.0893*.230*** *** (.0456) (.0401) (.0403) (.0402) ln_av_n_backward_citation.0268***.0165***.0319***.0130** (.00628) (.00473) (.00864) (.0066) ln_n_claims(initial).138*** ***.00153**.144*** *** (.00675) ( ) (.00638) ( ) (.00754) (.00118) (.0118) (.00149) ln_n_inventor ** ***.00372*** ** * (.00886) ( ) (.00804) ( ) (.00981) (.00134) (.0121) (.00153) ln_n_forward_citation *** * (.00642) ( ) (.00544) ( ) (.00697) (.00102) (.00957) (.00126) subappln_flg.0833*** *.0642*** ***.133*** *** (.0175) (.0018) (.0189) (.00215) (.0245) (.00379) (.0307) (.00404) US_Grant_flg ***.0251**.00762*** *** *** (.0110) (.00107) (.0119) (.00117) (.0153) (.00188) (.0282) (.00314) effective filing year yes yes yes yes yes yes yes yes 33_clasiffication yes yes yes yes yes yes yes yes applicant FE FE FE FE FE FE FE FE Observations 67,233 67,233 73,199 73,199 54,065 54,065 30,084 30,084 R-squared Number of patent owners Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.10 p. 27