Journal of Economic Behavior and Organization 166 (2019) 667–687

Contents lists available at ScienceDirect

Journal of Economic Behavior and Organization

journal homepage: www.elsevier.com/locate/jebo

The citation trap: Papers published at year-end receive

systematically fewer citations
Chao Ma a,b, Yiwei Li c, Feng Guo d,∗, Kao Si e
a
School of Public Health, Southeast University, No.87, Dingjiaqiao Road, Nanjing, 210009, China
b
School of Public Health, Yale University, U.S
c
Department of Marketing and International Business, SEK101, 1/F Simon and Eleanor Kwok Building, Lingnan University, Tuen Mun,
Hong Kong, China
d
School of Public Economics and Administration, Shanghai University of Finance and Economics, No.777, Guoding Road, Shanghai,
200433, China
e
Faculty of Business Administration, University of Macau, E22, Avenida da Universidade, Macau, Taipa, China

a r t i c l e i n f o a b s t r a c t

Article history: The present research reveals that academic papers published at year-end on average re-
Received 27 December 2018 ceive systematically fewer citations than papers published at other times in the year. Using
Revised 30 July 2019
more than 20 0,0 0 0 papers in economics published between 1956 and 2010, the results of
Accepted 9 August 2019
our analysis show that papers published between October and December on average get
Available online 17 August 2019
as much as 18.5% fewer citations than those published in the other months in the year. We
JEL classification: refer to this phenomenon as the citation trap as there is no evidence that papers published
A11 at different times in the year differ in their academic quality. We propose that the current
C21 effect could arise because of the time window options in most online academic search en-
D63 gines: the specific setting of those options leads papers published at year-end to appear
in the engines’ search results for a systematically shorter period of time as compared to
Keywords:
papers published at other months in the year. Our analysis reveals evidence that is con-
Citation
sistent with the proposed mechanism and that rules out several alternative explanations.
Publication timing
Year-end Implications of the current research for academia and possible solutions to mitigate the
Search engine citation trap are discussed.
© 2019 Elsevier B.V. All rights reserved.

1. Introduction

The number of citations an academic paper receives is a straightforward index of its impacts on subsequent research
and is thus often considered a reliable measure of the quality of the paper (Hamermesh et al., 1982; Hirsch, 2005; Smart
and Waldfogel, 1996). As a result, number of citations or indices based on citations (e.g., the H-index [Hirsch, 2005]) often
serves as a key determinant of many decisions in academia (Huang, 2015), including the hiring, pay, and promotion of
scholars (Diamond 1986; Ellison, 2013; Hilmer et al., 2015) and the conferment of scholarships and research funds (Berger,
2016; Garfield, 1999; Hamermesh et al., 1982; Smart and Waldfogel, 1996). As an example demonstrating the high impact
of number of citations, relevant authorities in the United Kingdom have recently announced their intention to transform
the Research Assessment Exercise (RAE) from a process that relies on peer reviews to one that is based on bibliometric

∗
Corresponding author.
E-mail addresses: [email protected], [email protected] (C. Ma), [email protected] (Y. Li), [email protected] (F. Guo),
[email protected] (K. Si).

https://doi.org/10.1016/j.jebo.2019.08.007
0167-2681/© 2019 Elsevier B.V. All rights reserved.
668 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687

data (Ellison, 2013). Recent evidence, however, suggests that the number of citations a paper receives could be affected by
factors exogenous to its academic quality. For example, the positions of papers within a journal could affect the frequency
with which they are cited by others (Coupé et al., 2010). Papers with a first author whose surname initial is earlier in
the alphabet get cited more frequently because they appear earlier in a reference list in many journals (Huang, 2015).
Furthermore, longer titles of research articles increase the chance that the papers be found by academic search engines
and thus lead to more citations of the papers (Guo et al., 2018).
In this research, we investigate the unexplored relationship between the number of citations that academic papers re-
ceive and their timing of publication in the year. Our findings indicate that papers published at year-end (e.g., October,
November, and December) on average get significantly fewer citations than those published at other months in the year.
Because the length of the review process and hence the exact timing of publication are beyond the control of most scholars,
it is highly unlikely that the phenomenon we document here reflects a relationship between the academic quality of the
papers and their timing of publication in the year. Therefore, timing of publication in the year could constitute a systematic
bias when it comes to recognizing the quality of academic papers and the professional attainments of scholars. We hence
refer to the current effect as the citation trap.
We suggest that the citation trap may stem from the time window options in most online academic search engines
(e.g., Web of Science, Jstor, ScienceDirect, Springer, and Google Scholar). Online search engines constitute a major avenue via
which scholars seek references for their research and thus have important influences on academic citations (Guo et al.,
2018; Kousha and Thelwall, 2007). To narrow the returned search results, most search engines offer their users the option
to search articles published within certain time windows, usually in units of years. For example, Springer allows their users
to search papers published between two particular years (e.g., between 2015 and 2019); Google Scholar allows their users
to search papers published after a particular year (e.g., since 2015; see Appendix A for more examples). Importantly, when
users of search engines choose to impose time windows on their searches, the engines will return relevant articles published
on or after the first day of the beginning year. For example, if a scholar searches for papers published “between 2015 and
2019” or “since 2015” on July 1st, 2019, the search engines will return papers that are published on or after January 1st,
2015, rather than on or after July 1st, 2015.
To see the effect of this specific setting on papers published at year-end (versus those published at other times in the
year), suppose that a scholar always searches for papers published in the recent five years (e.g., in 2019 his searches shall be
“between 2015 and 2019” or “since 2015”). Then a paper published in December 2015 shall appear in the scholar’s search
results during the period between December 2015 and December 2019 (49 months) while a paper published in January
2015 shall appear in the scholar’s search results during the period between January 2015 and December 2019 (60 months).
Therefore, when the time window options in the search engines are used, papers published at year-end suffer a disadvantage
relative to papers published at other times in the year in terms of the duration in which they appear in the engines’ search
results. Importantly, this disadvantage does not require that the year-end papers be published temporally after the non-year-
end papers. For example, a paper published in January 2016 shall also appear in the above scholar’s search results for a total
of 60 months (between January 2016 and December 2020). More generally, if the time window is set to be n years (n equals
to 5 in the above example), then papers published in the mth (m ∈ [1, 12]) month in the year shall appear in the engines’
search results for 12(n – 1) + (12 – m + 1) = 12n – m + 1 months. It is clear that as m increases (i.e., publishing papers in
later months in the year), the length of time during which the papers appear in the engines’ search results decreases. We
propose that this disadvantage of papers published at year-end with regard to the duration in which they appear in the
engines’ search results could potentially drive the citation trap.
In the current research, we document the citation trap and provide evidence that supports the proposed mechanism.
Using more than 20 0,0 0 0 SSCI papers in economics, we show that papers published between October and December on av-
erage get as much as 18.5% fewer citations than those published in the other months in the year. We demonstrate the effect
after controlling for the influences of several attributes of the paper, including paper length, number of authors, number
of references cited, and title length, and we show that the results remain robust across different model specifications and
identification strategies. Moreover, our analyses provide evidence that supports the time window options in search engines
as the underlying mechanism of the current effect. First, our results show that there is a mitigating effect of journal quality
on the citation trap. Specifically, the effect is less pronounced for papers published in top-tier journals than for papers pub-
lished in lower-tier journals. Because top-tier journals have much greater and more engaged readership, scholars rely much
less on search engines to access papers published in those journals. Therefore, if the time window options in search engines
drive the citation trap, the effect should be attenuated for papers published in top-tier journals.
Second, the results show that there is a mitigating effect of the length of time that has passed since the paper was pub-
lished. The effect is more pronounced for papers published in the more recent years when the use of online academic search
engines became prevalent. Importantly, our mechanism does not require that the papers be published after the emergence
of the online engines. That is, the citation trap might affect papers as long as they can be found using the time window
options in online search engines. Nevertheless, the moderating effect of time is consistent with the current mechanism be-
cause the purpose of using the time window options is to narrow the returned search results and therefore the majority of
the engine users should choose to impose relatively short time windows in their searches (i.e., search for papers that were
published in the recent past). Moreover, because papers get more citations over time, it is also consistent with the current
mechanism that the effect of the citation trap becomes proportionally smaller in the long term. Finally, we provide further
evidence that rules out several alternative explanations of the current effect. Our results suggest that the effect is unlikely
 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687 669

Fig. 1. The total number of papers published in each calendar month.

Notes: The dataset was retrieved from the Web of Science and consists of papers in economics published between 1956 and 2010.

Fig. 2. The mean and the median of number of citations of papers published in each calendar month.
Notes: The dataset was retrieved from the Web of Science and consists of papers in economics published between 1956 and 2010.

to arise because of (1) strategic submissions by high ability scholars, (2) plausible influences of journal editors at year-end,
or (3) the ordering of papers by the search engines.
The reminder of the paper is organized as follows. We describe the data and specify our empirical model in Section 2.
We report our results and provide evidence that supports our proposed mechanism and rules out alternative accounts in
Section 3. Finally, we conclude in Section 4.

2. Data and model specification

We retrieved information of published papers in economics from the Web of Science (WoS) database. The papers were
published between 1956 and 2010. We excluded from our analyses papers of which the publication date could not be iden-
tified and papers that were published in special issues of the journals, in which the overall quality of the papers is often a
concern (Coupé et al., 2010; Hudson, 2007). The final dataset consists of 208,977 papers.
Fig. 1 shows the total number of papers published in each calendar month over the years. While substantially more
papers were published in March, June, September, and December due to the arrangement of quarterly journals, there was
no substantial difference in the number of papers published in the other months. For each paper in our dataset, we recorded
its total number of citations received before December 2012. The mean and the median number of citations of papers in
our dataset were 18.12 (SD = 69.83) and 5, respectively. Fig. 2 shows the mean and the median of number of citations of
papers published in each calendar month over the years. As can be seen from the figure, papers published at year-end
(October to December), particularly those published in December, had substantially fewer citations than those published at
other months in the year. Furthermore, Fig. 3 depicts the distributions of number of citations of papers published between
October and December and of papers published in the other months. As can be seen, papers published at year-end (between
670 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687

Fig. 3. Distributions of number of citations of papers published between October and December and of papers published between January and September.
Notes: The dataset was retrieved from the Web of Science and consists of papers in economics published between 1956 and 2010.

October and December) had noticeably higher density values in regions of smaller numbers of citations. The results of Epps-
Singleton two-sample test (Goerg and Kaiser, 2009) confirmed that the two distributions were significantly different, p <
0.001. Therefore, the findings from Figs. 2 and 3 suggested that papers published at year-end overall received fewer citations
than papers published at other times in the year.
To formally examine the relationship between the number of citations of academic papers and their timing of publication
in the year, we constructed the following benchmark model:

ln(citation )i jt = β yr _endi jt + ∅Xi jt + yeart + journal j + εi jt (1)

where ln(citation)ijt is the natural logarithm of the number of citations of paper i published in journal j in year t (we added
0.0 0 01 to the number of citations of each paper before taking the logarithm transformation to include papers with zero
citation in our model). yr _endi jt is a dummy variable that indicates whether the paper was published at year-end. In our
benchmark model, year-end was defined to include October, November, and December. Therefore, yr _endi jt equals 1 if the
corresponding paper was published between October and December and it equals 0 if the paper was published in the other
months in the year. Among the papers in our dataset, 53,707 (25.7%) of them were published in year-end according to
the current definition. yeart and journalj represent the fixed effects of the year and the journal in which the paper was
published, respectively. ɛijt is the error term.
We included in our model a set of control variables X that may correlate with the number of citations a paper receives.
These included (1) the number of pages of the paper (Card and Della Vigna, 2014; Vieira, 2008); (2) the number of authors
of the paper (Di Vaio et al., 2012; Freeman and Huang, 2015; Sauer, 1988; Vieira, 2008); (3) the alphabetical order of the
first author’s surname initial (Huang, 2015)–we used a dummy variable to represent the place of the first author’s surname
initial in the alphabet: the variable equals 1 if the surname initial is between A and M in the alphabet (146,075 [69.9%] of
the papers in our dataset fell in this category) and it equals 0 if the surname initial is between N and Z in the alphabet;
(4) the number of references that were cited in the paper (Vieira and Gomes 2010; Webster et al., 2009); (5) a dummy
variable indicating whether the paper was a leading article (Berger, 2016; Coupé et al., 2010; Hudson, 2007; Pinkowitz,
2002): the variable equals 1 if it was and 0 if it was not (24,868 [11.9%] of the papers in our dataset were leading articles);
(6) the number of letters in the paper’s title (Bramoullé and Ductor, 2018; Guo et al., 2018; Letchford et al., 2015); and
(7) the number of papers in the same issue of the journal in which the paper was published (King, 2004). See Table 1
for a summary of the descriptive statistics of these control variables. In Appendix B, we demonstrate the means of these
variables for papers published in each calendar month. Importantly, there was no substantial difference between the means
in different calendar months.
In all cases we were interested in the estimation of β , which represents the effect of publishing at year-end relative to
other times in the year on the number of citations a paper receives. We reported robust standard errors clustered at the
journal-year level.
 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687 671

Table 1
Descriptive statistics for some of the control variables in the model.

Variables Obs Mean SD Median Min Max

Number of pages 208,977 17.105 8.736 16 3 100

Number of authors 208,977 1.726 0.895 2 1 10
Number of references cited 208,977 26.016 18.721 23 0 680
Number of letters in title 208,977 9.511 3.777 9 1 45
Number of papers in the same issue 208,977 11.097 7.710 9 1 122
Author team ability 208,977 4.625 6.824 2 0 188
Author team degree of multidiscipline 208,977 2.536 2.651 2 0 33

Notes: Number of pages is the number of pages of the paper; Number of authors is the number of authors
of the paper; Number of references cited is the number of references that were cited in the paper; Number
of letters in title is the number of letters in the title of the paper; Number of papers in the same issue is
the number of papers in the same issue of the journal in which the paper was published; Author team
ability is the total number of academic papers published by all the members in the author team of the
paper during the last five years before the paper was published; Author team degree of multidiscipline
is the sum of the number of disciplines in which each member in the author team of the paper had
published at least one paper during the last five years before the paper was published.

Table 2
Model estimation results.

Variables (1) (2) (3) (4) (5) (6)

Year-end −0.242∗∗∗ −0.255∗∗∗ −0.268∗∗∗ −0.191∗∗∗ −0.189∗∗∗ −0.185∗∗∗

(0.020) (0.019) (0.044) (0.033) (0.043) (0.021)
Number of pages 0.0689∗∗∗ 0.0664∗∗∗ 0.0558∗∗∗ 0.0451∗∗∗ 0.0563∗∗∗
(0.001) (0.006) (0.003) (0.004) (0.002)
Number of authors 0.456∗∗∗ 0.552∗∗∗ 0.314∗∗∗ 0.337∗∗∗ 0.306∗∗∗
(0.010) (0.018) (0.021) (0.018) (0.010)
Place of surname initial 0.115∗∗∗ 0.126∗∗∗ 0.0205 0.0255 0.0347∗∗
(0.019) (0.027) (0.020) (0.021) (0.017)
Number of references cited 0.0315∗∗∗ 0.0381∗∗∗ 0.0244∗∗∗ 0.0279∗∗∗ 0.0251∗∗∗
(0.001) (0.001) (0.002) (0.002) (0.001)
Leading article 0.164∗∗∗ 0.135∗∗∗ 0.423∗∗∗ 0.417∗∗∗ 0.410∗∗∗
(0.027) (0.038) (0.029) (0.030) (0.022)
Number of letters in title −0.0272∗∗∗ −0.0247∗∗∗ 0.00802∗ 0.00148 −0.00495∗∗
(0.002) (0.003) (0.004) (0.003) (0.002)
Number of papers in the same issue 0.0269∗∗∗ 0.0361∗∗∗ −0.0279∗∗∗ −0.0167∗ −0.0250∗∗∗
(0.001) (0.011) (0.006) (0.008) (0.004)
Year fixed effects NO NO YES NO YES NO
Journal fixed effects NO NO NO YES YES NO
Journal-year fixed effects NO NO NO NO NO YES
N 208,977 208,977 208,977 208,977 208,977 208,946
R2 0.000682 0.0777 0.119 0.269 0.298 0.324

Notes: The dependent variable in the models is the natural logarithm of the sum of the number of citations of the paper and the number 0.0 0 01. Year-end
is a dummy variable that equals 1 if the paper was published between October and December and equals 0 if otherwise; Number of pages is the number
of pages of the paper; Number of authors is the number of authors of the paper; Place of surname initial is a dummy variable that equals 1 if the surname
initial of the first author of the paper is between A and M in the alphabet and equals 0 if the surname initial of the first author of the paper is between N
and Z in the alphabet; Number of references cited is the number of references that were cited in the paper; Leading article is a dummy variable that equals
1 if the paper was a leading article in the issue and equals 0 if was not; Number of letters in title is the number of letters in the title of the paper; Number
of papers in the same issue is the number of papers in the same issue of the journal in which the paper was published. The standard errors in parentheses
in column (3), (4), (5), and (6) are clustered at their corresponding fixed-effect level(s). ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1.

3. Results

3.1. Benchmark models

We presented the estimation results of our benchmark model and several alternative model specifications in Table 2.
Column 1 shows the results when only the yr _endi jt variable was included in the model. The regression coefficient of the
variable was negative and significant at the 0.01 level, providing initial evidence that publishing at year-end has negative
effects on the number of citations that the papers receive. The magnitude of this coefficient did not change significantly
when we added the control variables and the fixed effects into the model (see Columns 2 to 6, Table 2). In Column 6,
we controlled for systematic differences at the journal-year level by adding the interaction effect between the yeart and
the journalj variables in the model. The model specification in Column 6 had the largest adjusted R2 , suggesting that the
explanatory power of the model was enhanced by adding the journal-year fixed effect. Therefore, we adopted the estimates
in Column 6 as the benchmark results, which suggested that, ceteris paribus, the number of citations of papers published
at year-end was on average 18.5% smaller than that of papers published in the other months in the year. The signs of the
672 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687

Table 3
Model estimation results with data trimming and alternative model specifications.

Variables (1) (2) (3) (4) (5)

Excluding Zero Excluding Top 5% Excluding December Nb-reg Inv-sine

Year-end −0.0651∗∗∗ −0.186∗∗∗ −0.157∗∗∗ −0.0907∗∗∗ −0.0788∗∗∗

(0.007) (0.021) (0.025) (0.009) (0.007)
Control variables YES YES YES YES YES
Journal-year fixed effects YES YES YES YES YES
N 180,644 198,593 185,469 208,977 208,946
R2 0.383 0.296 0.324 N/A 0.444

Notes: The dependent variable in Column 1 is the natural logarithm of the number of citations of the paper and thereby
Column 1 shows the model estimation results when papers with zero citation were excluded. The dependent variable in
Columns 2 to 4 is the natural logarithm of the sum of the number of citations of the paper and the number 0.0 0 01.
Column 2 shows the results when papers with citation numbers that were at the 95th percentile were excluded. Column 3
shows the results when papers published in December were excluded. Column 4 shows the results of the negative binomial
model. Column 5 shows the results when the dependent variable is the number of citations of the paper transformed by the
inverse hyperbolic sine function. Year-end is a dummy variable that equals 1 if the paper was published between October
and December and equals 0 if otherwise. The control variables include all the seven variables described in Section 2. The
standard errors in parentheses are clustered at the journal-year level. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1.

coefficients of the control variables were consistent with implications of previous research (Bramoullé and Ductor, 2018;
Card and Della Vigna, 2014; Coupé et al., 2010; Di Vaio et al., 2012; Freeman and Huang, 2015; Hudson, 2007; Vieira, 2008).

3.2. Robustness checks

3.2.1. Data trimming and alternative model specifications

We conducted several robustness checks to provide further evidence of the validity of the current effect. First, we ex-
cluded papers that had zero or extremely large numbers of citations and ran the benchmark model (model specification as
shown in Column 6, Table 2) again. Column 1 and Column 2 of Table 3 list the results when papers with zero citation and
when papers with citation numbers that were at the 95th percentile in our dataset were excluded, respectively. The coeffi-
cients of the yr _endi jt variable remained negative and significant in both cases. The results thus suggested that the citation
trap was not driven by papers with extreme numbers of citations. Moreover, it was clear from Fig. 2 that papers published
in December on average received particularly fewer citations. To check if the year-end effect still holds if the impacts of
those papers are removed, we ran the benchmark model with all papers published in December being excluded from our
analysis. The results (Column 3, Table 3) confirmed that the citation trap remained robust.
Further, we used alternative model specifications to account for the non-linearity inherent in our data. We specified a
negative binomial regression to cater for the count data nature of number of citations. In addition, to examine the robust-
ness of our findings under alternative transformation, we applied the inverse hyperbolic sine transformation to the number
of citations of the papers. Column 4 and Column 5 of Table 3 present the estimation results of the above two model spec-
ifications respectively. The coefficients of the yr _endi jt variable were negative and significant in both columns, therefore
suggesting that the current effect was robust under these alternative model specifications.

3.2.2. Alternative definitions of year-end

To eliminate the concern that the citation trap is specific to our definition of year-end in the benchmark model (i.e.,
October to December), we varied the operationalization of the yr _endi jt variable and re-ran our benchmark model. Specifi-
cally, we adopted the following alternative definitions of year-end to include (1) December, (2) November and December, (3)
September to December, (4) August to December, and (5) July to December, respectively. Importantly, the estimation results
(see Table 4) of our model with these alternative definitions of year-end confirmed that the citation trap was robust across
all the definitions.

3.2.3. Alternative definitions of year

As discussed above, the citation trap does not arise merely because the year-end papers were published temporarily after
the non-year-end papers. To more clearly address this concern, we re-defined a year in our model to enable us to compare
the effect of publishing papers at year-end to that of publishing papers at later times (i.e., in the next year). Specifically, for
example, under the December-Year definition, each year starts from December and ends at November of the next year. When
we used this definition of year in our model estimation, thereby, the yr _endi jt variable equals 1 if the paper was published
in December in one year or between October and November in the next year. We used several alternative definitions of
year and re-ran our benchmark model. The results (see Table 5) indicated that the current effect remained negative and
significant when we compared the number of citations of papers published at year-end with that of papers published in
the next year. For example, the results in Column 3 of Table 5 suggested that the average number of citations of papers
published between October and December was 6.4% smaller than that of papers published between January and September
in the next year.
 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687 673

Table 4
Model estimation results with alternative definitions of year-end.

Variables (1) (2) (3) (4) (5)

Dec Nov-Dec Sep-Dec Aug-Dec Jul-Dec

Year-end −0.227∗∗∗ −0.215∗∗∗ −0.191∗∗∗ −0.197∗∗∗ −0.201∗∗∗

(0.033) (0.025) (0.019) (0.018) (0.017)
Control variables YES YES YES YES YES
Journal-year fixed effects YES YES YES YES YES
N 208,946 208,946 208,946 208,946 208,946
R2 0.324 0.324 0.324 0.324 0.324

Notes: The dependent variable in the models is the natural logarithm of the sum of the number
of citations of the paper and the number 0.0 0 01. We ran our benchmark model with alternative
definitions of year-end. In Column 1 year-end was defined to include December; In Column 2
year-end was defined to include November and December; In Column 3 year-end was defined to
include September to December; In Column 4 year-end was defined to include August to Decem-
ber; In Column 5 year-end was defined to include July to December. The control variables include
all the seven variables described in Section 2. The standard errors in parentheses are clustered at
the journal-year level. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1.

Table 5
Model estimation results with alternative definitions of year.

Variables (1) (2) (3) (4) (5) (6)

Dec-Year Nov-Year Oct-Year Sep-Year Aug-Year Jul-Year

Year-end −0.135∗∗∗ −0.101∗∗∗ −0.0638∗∗∗ −0.0807∗∗∗ −0.0944∗∗∗ −0.101∗∗∗

(0.021) (0.021) (0.021) (0.021) (0.021) (0.020)
Control variables YES YES YES YES YES YES
Journal-year fixed effects YES YES YES YES YES YES
N 208,946 208,950 208,953 208,951 208,951 208,947
R2 0.325 0.325 0.325 0.325 0.324 0.325

Notes: The dependent variable in the models is the natural logarithm of the sum of the number of citations of
the paper and the number 0.0 0 01. We ran our benchmark model with alternative definitions of year. In Column 1,
each year was defined as starting from December and ending at November in the next year, and year-end included
December in one year and October and November in the next year. In Column 2, each year was defined as starting
from November and ending at December in the next year, and year-end included November and December in one
year and October in the next year. In Column 3, each year was defined as starting from October and ending at
September in the next year. In Column 4, each year was defined as starting from September and ending at August
in the next year. In Column 5, each year was defined as starting from August and ending at July in the next year.
In Column 6, each year was defined as starting from July and ending at June in the next year. In Columns 3 to
6, year-end included October to December in the earlier year. The control variables include all the seven variables
described in Section 2. The standard errors in parentheses are clustered at the journal-year level. ∗∗∗ p < 0.01, ∗∗ p
< 0.05, ∗ p < 0.1.

Furthermore, we took an alternative approach to investigate the current effect. Specifically, we compared the effect of
publishing papers in January with that of publishing papers in each of the other months in the year. We created dummy
variables to represent the effect of publishing in each month between February and December relative to that in January
and we examined the effects under different definitions of year. The results are depicted in Table 6. For example, under the
normal definition of year (i.e., starting from January and ending at December), the results (Column 1, Table 6) suggested that
the number of citations of papers published between February and December decreased almost monotonically as compared
to that of papers published in January. Similar patterns of results were obtained when we used other definitions of year
(Columns 2 to 7, Table 6). Further, the results in Columns 2 to 7 confirmed again that papers published at the end of a year
received significantly fewer citations than papers published in January of the next year. To conclude, our results confirmed
that the citation trap remained significant under a variety of robustness checks and thus provided strong support to the
current effect.

3.3. Evidence supporting the proposed mechanism

In this section we provide evidence that is consistent with our proposed mechanism. According to the current mecha-
nism, we made two predictions. First, we predicted that the observed effect should be less pronounced for paper published
in top-tier journals than for papers published in lower-tier journals. This is because scholars rely less on search engines
for knowledge of or access to papers published in the former than those published in the latter. To test this prediction, we
created a dummy variable topj . It equals 1 if journal j was one of the top-tier journals in economics1 and it equals 0 if

1
The top-tier journals include American Economic Review, Econometrica, Journal of Political Economy, Quarterly Journal of Economics, Review of Economic
Studies, Economic Journal, Games and Economic Behavior, International Economic Review, Journal of Econometrics, Journal of Economic Theory, Journal of Finance,
674 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687

Table 6
Model estimation results with alternative definitions of year and dummy variables representing the effect of publishing in each
month between February and December relative to that in January.

Variables (1) (2) (3) (4) (5) (6) (7)

Jan-Year Dec-Year Nov-Year Oct-Year Sep-Year Aug-Year Jul-Year

Feb 0.00298 −0.00161 0.0151 0.0109 0.00452 0.0159 0.0161

(0.047) (0.046) (0.046) (0.045) (0.045) (0.045) (0.045)
Mar −0.0877∗∗ −0.0813∗ −0.0799∗ −0.0859∗∗ −0.0756∗ −0.0744∗ −0.0754∗
(0.043) (0.043) (0.042) (0.043) (0.042) (0.042) (0.043)
Apr −0.0572 −0.0570 −0.0527 −0.0483 −0.0499 −0.0507 −0.0443
(0.043) (0.043) (0.043) (0.042) (0.043) (0.043) (0.043)
May −0.115∗∗ −0.112∗∗ −0.0983∗∗ −0.101∗∗ −0.107∗∗ −0.0890∗∗ −0.0904∗∗
(0.046) (0.046) (0.045) (0.045) (0.045) (0.045) (0.045)
Jun −0.158∗∗∗ −0.151∗∗∗ −0.147∗∗∗ −0.152∗∗∗ −0.154∗∗∗ −0.158∗∗∗ −0.156∗∗∗
(0.044) (0.043) (0.043) (0.043) (0.044) (0.043) (0.044)
Jul −0.194∗∗∗ −0.192∗∗∗ −0.188∗∗∗ −0.186∗∗∗ −0.184∗∗∗ −0.184∗∗∗ −0.0639
(0.042) (0.043) (0.043) (0.043) (0.043) (0.043) (0.042)
Aug −0.233∗∗∗ −0.235∗∗∗ −0.218∗∗∗ −0.217∗∗∗ −0.220∗∗∗ −0.107∗∗ −0.108∗∗
(0.047) (0.047) (0.047) (0.047) (0.047) (0.047) (0.047)
Sep −0.276∗∗∗ −0.270∗∗∗ −0.266∗∗∗ −0.270∗∗∗ −0.137∗∗∗ −0.136∗∗∗ −0.136∗∗∗
(0.044) (0.043) (0.043) (0.043) (0.044) (0.044) (0.043)
Oct −0.204∗∗∗ −0.200∗∗∗ −0.201∗∗∗ −0.0690 −0.0731∗ −0.0758∗ −0.0652
(0.044) (0.044) (0.043) (0.043) (0.043) (0.043) (0.043)
Nov −0.319∗∗∗ −0.319∗∗∗ −0.187∗∗∗ −0.185∗∗∗ −0.191∗∗∗ −0.185∗∗∗ −0.188∗∗∗
(0.052) (0.051) (0.050) (0.051) (0.050) (0.050) (0.050)
Dec −0.386∗∗∗ −0.261∗∗∗ −0.255∗∗∗ −0.262∗∗∗ −0.258∗∗∗ −0.259∗∗∗ −0.253∗∗∗
(0.047) (0.046) (0.046) (0.046) (0.046) (0.046) (0.045)
Control variables YES YES YES YES YES YES YES
Journal-year fixed effects YES YES YES YES YES YES YES
N 208,946 208,946 208,950 208,953 208,951 208,951 208,947
R2 0.324 0.325 0.325 0.325 0.325 0.325 0.325

Notes: The dependent variable in the models is the natural logarithm of the sum of the number of citations of the paper and
the number 0.0 0 01. We compared the effect of publishing papers in January with that of publishing papers in each of the
other months under alternative definitions of year. In Column 1, each year was defined normally as starting from January and
ending at December. In Column 2, each year was defined as starting from December and ending at November in the next year.
In Column 3, each year was defined as starting from November and ending at October in the next year. In Column 4, each
year was defined as starting from October and ending at September in the next year. In Column 5, each year was defined as
starting from September and ending at August in the next year. In Column 6, each year was defined as starting from August and
ending at July in the next year. In Column 7, each year was defined as starting from July and ending at June in the next year.
The dummy variables Feb to Dec represent the effect of publishing in each month between February and December relative to
that in January, respectively. The control variables include all the seven variables described in Section 2. The standard errors in
parentheses are clustered at the journal-year level. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1.

otherwise. We added the topj variable and its interaction with the yr _endi jt variable into the benchmark model and ran the
model again. The results are shown in Column 1, Table 7. Importantly, supporting our prediction, the interaction between
the topj and the yr _endi jt variables was positive and significant, suggesting that the negative effect of publishing at year-end
was significantly attenuated for papers published in top-tier journals.
Second, to the extent that the purpose of using the time window options is to narrow the returned search results, the
majority of the engine users should choose to use relative short time windows in their searches. Therefore, per the current
mechanism, we predicted that the citation trap should be more likely to affect papers that were published in the recent
past than those that were published in the distant past. Furthermore, the moderating effect of time could also arise because
papers get more citations over time, and therefore the difference in number of citations as induced by the citation trap
becomes proportionally smaller in the long term. To examine the predicted effect of the length of time that has passed since
the paper was published, we divided the papers in our dataset into five sub-samples according to their time of publication
(between 1956 and 1970, between 1971 and 1980, between 1981 and 1990, between 1991 and 20 0 0, and between 2001
and 2010) and we ran our benchmark model in each sub-sample. In addition, we created a variable trendt to indicate the
time when the paper was published. It equals 1 for papers in the most distant sub-sample (i.e., published between 1956
and 1970) and 5 for papers in the most recent sub-sample (i.e., published between 2001 and 2010). We added the variable
and its interaction with the yr _endi jt variable in our benchmark model and we ran the model using the whole sample. The
results are shown in Table 7. The results in Columns 2 to 6 showed that the negative year-end effect was significant in the
three sub-samples with the most recent papers but became non-significant in the other two sub-samples with older papers.
Further, the results in Column 7 confirmed that the interaction between the trendt and yr _endi jt variables was negative and

Journal of International Economics, Journal of Labor Economics, Journal of Monetary Economics, Journal of Public Economics, Journal of the European Economic
Association, Rand Journal of Economics, Review of Economics and Statistics, and Theoretical Economics.
 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687 675

Table 7
Model estimation results showing the moderating effect of journal quality and time.

Variables (1) (2) (3) (4) (5) (6) (7)

All 56–70 71–80 81–90 91–00 01–10 All

Year-end −0.205∗∗∗ −0.0290 −0.0611 −0.157∗∗∗ −0.0807∗∗ −0.291∗∗∗ 0.0616

(0.024) (0.107) (0.080) (0.054) (0.036) (0.031) (0.082)
Year-end × Top 0.118∗∗∗
(0.044)
Year-end × Trend −0.0617∗∗∗
(0.019)
Control variables YES YES YES YES YES YES YES
Journal-year fixed effects YES YES YES YES YES YES YES
N 208,946 9609 15,248 33,236 60,129 90,724 208,946
R2 0.324 0.312 0.336 0.365 0.332 0.280 0.324

Notes: The dependent variable in the models is the natural logarithm of the sum of the number of citations of the paper and
the number 0.0 0 01. Year-end is a dummy variable that equals 1 if the paper was published between October and December
and equals 0 if otherwise. In Column 1, Top is a dummy variable that equals 1 if the paper was published in one of the
top-tier journals in economics and equals 0 if otherwise. In Columns 2 to 6, we divided the papers in our dataset into five
sub-samples according to their time of publication and we ran the benchmark model in each sub-sample. In Column 7, Trend
is a variable indicating the time when the paper was published. It equals 1 for papers in the most distant sub-sample (i.e.,
published between 1956 and 1970) and 5 for papers in the most recent sub-sample (i.e., published between 2001 and 2010).
The control variables include all the seven variables described in Section 2. The standard errors in parentheses are clustered
at the journal-year level. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1.

significant, indicating that the current effect was weaker for papers published in the more distant past. Together, the results
provided support to our predictions and were thus consistent with the proposed mechanism.

3.4. Alternative accounts

3.4.1. Strategic submissions by high ability scholars

One alternative explanation of the current effect is that some experienced or high ability scholars might somehow be
aware that papers published at year-end tend to receive fewer citations and they try to avoid publishing at year-end by
strategically submitting their papers at certain times in the year (taking into account the average duration of the review
process). Although we think that such strategic submission is unlikely and difficult to achieve, it is important that we rule
out this possibility because it suggests that the observed difference in number of citations could be due to difference in the
academic quality of the papers.
We therefore took several measures to examine the validity of this alternative explanation. First, for each paper in our
dataset, we measured the ability of its author team by calculating the total number of academic papers published by all
the members2 during the last five years before the paper was published. We also measured the degree to which the author
team was multidisciplinary. Specifically, we counted the number of disciplines (among the 95 disciplines as specified by
WoS) in which an author had published at least one paper during the last five years before the paper was published. We
did this for each author of the paper and we calculated the sum for the author team. We re-ran our benchmark model
with these two variables added in the model (see Table 1 for the descriptive statistics of these two variables). The results
are shown in Column 1, Table 8. Not surprisingly, the results showed that both ability and degree of multidiscipline of the
author team positively influenced the number of citations of the papers. Importantly, however, the effect of the yr _endi jt
variable remained significant after author team ability and degree of multidiscipline were controlled for.
Second, we looked at papers published by an elite group of authors—the Nobel laureates in Economics (before 2019). We
retrieved information for all the papers involving at least one of the laureates as a co-author and we depicted the number of
papers published in each calendar month in Fig. 4. As can be seen, the number of published papers by the Nobel laureates
was not substantially smaller at year-end (e.g., between October and December) than at other times in the year. Further, the
results of our analysis using papers in our original dataset showed that the proportion of papers published at year-end was
not significantly different between those with at least one Nobel laureate as a co-author (25%; 658/2626) and those without
(26%; 53,089/206,351), χ 2 (1) < 1. The results therefore indicated that at least the Nobel laureates, a group of scholars with
recognizably high ability and rich experience in academic publications, did not try to strategically submit their papers (or
not successful in doing so) as to avoid publishing at year-end.
Finally, we investigated the prevalence rate of strategic submission by conducting an online survey among scholars. The
results confirmed that strategic submission in terms of timing of publication was neither common among scholars nor
was it associated with their ability or experience. Please see Appendix C for results and discussions of the survey and
Appendix D for the instructions and questions used in the survey. Together, the results across different methods converged
to the conclusion that strategic submission is unlikely to cause the observed effect.

2
A pair of names that have identical surnames and first name initials are identified under one unique author.
676 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687

Table 8
Model estimation results for testing of alternative accounts.

Variables (1) (2) (3)

∗∗∗ ∗∗∗
Year-end −0.181 −0.922 −0.341∗ ∗ ∗
(0.021) (0.090) (0.037)
Author team ability 0.0136∗∗∗ 0.0117∗∗∗ 0.0117∗∗∗
(0.002) (0.002) (0.002)
Author team degree of multidiscipline 0.0504∗∗∗ 0.0294∗∗∗ 0.0296∗∗∗
(0.004) (0.007) (0.007)
Control variables YES YES YES
Journal-year fixed effects YES YES YES
Cragg-Donald Wald F 13,000
N 208,946 69,281 69,281
R2 0.326 0.268 0.272

Notes: The dependent variable in the models is the natural logarithm of the sum
of the number of citations of the paper and the number 0.0 0 01. Column 1 shows
the results when author team ability and degree of multidiscipline were controlled
for. Column 2 shows the results when the IV strategy was employed via a two-stage
least square estimation. Column 3 shows the results of the OLS counterpart that cor-
responds to the IV estimates. As data from Google Trends are available after 2004,
in Column 2 and Column 3 we only included data between 2004 and 2010 and the
sample size of the IV analysis and its corresponding OLS counterpart was reduced
to 69,281. Year-end is a dummy variable that equals 1 if the paper was published
between October and December and equals 0 if otherwise; Author team ability is the
total number of academic papers published by all the members in the author team
of the paper during the last five years before the paper was published; Author team
degree of multidiscipline is the sum of the number of disciplines in which each mem-
ber in the author team of the paper had published at least one paper during the last
five years before the paper was published. The control variables include all the seven
variables described in Section 2. The standard errors in parentheses are clustered at
the journal-year level. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1.

Fig. 4. Number of papers published involving at least one of the Nobel laureates in Economics (before 2019) as a co-author in each calendar month. .

3.4.2. Influences of journal editors

Another possibility is that the observed effect could arise because of possible influences of journal editors at year-end.
For instance, during year-end editors may be more likely to make lenient decisions and accept papers with less rigor. Al-
ternatively, editors might try to exploit the yearly quota for pages as imposed by publishers when approaching year-end
and thereby publish more papers. These factors could possibly cause papers that are published at year-end to have fewer
citations. In our benchmark model we have controlled for the effect of number of papers in the same issue of the journal
and indeed we found a negative effect of this variable on the number of citations received. To further eliminate the concern
that the observed effect might be due to influences of journal editors at year-end, we performed an instrumental variable
(IV) analysis to tackle potential endogeneity issues arising from journal editors.
We utilized data from Google Trends for searches for the keyword “spring festival” (the Chinese new year) as the extra
instrument (see Appendix E for a depiction of the data from Google Trends). The basic rationales of using the current instru-
ment are twofold. First, the spring festival usually takes place in the early months of a year (mostly in January and February)
and thus the search volumes for the keyword were negatively correlated with the yr _endi jt variable, r = −0.41, p < 0.001.
 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687 677

Second, the spring festival is highly unlikely to have a general effect on the decisions of journal editors of the papers in our
dataset.
A first stage estimation was implemented to investigate the correlation between the yr _endi jt variable and all exogenous
instruments. The Cragg-Donald Wald F statistic value was 13,0 0 0, which was greater than 10, suggesting that the relevance
condition was fulfilled. The estimates of the IV analysis are presented in Column 2 of Table 8 and the results of the OLS
counterpart that corresponds to the IV estimates are shown in Column 3 of Table 8. The results of the two columns showed
that the coefficient of the yr _endi jt variable was negative and significant with both the IV estimates and its OLS counterpart.3

3.4.3. The ordering of papers by the search engines

Finally, the negative effect of publishing at year-end on number of citations received could be due to the order by which
papers are shown in the search results of the engines. The default setting of most search engines arranges the papers in
terms of their impacts, with papers having greater numbers of citations appearing at the top. Therefore, papers published
earlier in the year may initially gain greater visibility (i.e., relatively larger number of citations) than papers published later
in the year, and this difference could be reinforced and enlarged by the ordering of papers by the search engines. However,
our findings suggest that the current effect is unlikely to be driven by this ordering effect. First, we showed that the negative
year-end effect remained robust even when we compared the effect of publishing at year-end with that of publishing in the
early months in the next year (see Tables 5 and 6). Second, if it is the ordering effect that drives and reinforces the citation
trap, then we should observe that the effect becomes larger as the length of time that has passed since the paper was
published increases. This, however, is diametrically opposite to what we have found (see Columns 2 to 7, Table 7).

4. Conclusion

In this research we identify the citation trap: papers published at year-end receive systematically fewer citations than
papers published at other months in the year. Using a large sample of more than 20 0,0 0 0 papers, the results of our analyses
confirm that papers published between October and December get significantly fewer citations than those published in the
other months in the year. To the best of our knowledge, we are among the first to investigate the relationship between the
number of citations received by academic papers and their timing of publication in the year.
We argue that the current effect could potentially arise because of the time window options in online academic search
engines and we provide evidence that is consistent with our proposed mechanism. Our results show that the effect is less
pronounced for papers published in top-tier journals to which scholars rely much less on search engines to access. The
results further show that the effect is more pronounced for papers published in the more recent years during which the use
of online search engines became prevalent. In addition, we rule out several alternative explanations of the current effect. Our
results suggest that the citation trap is unlikely to be driven by strategic submissions by high ability scholars, by possible
influences of journal editors, or by the ordering of papers by the search engines.
The current research has important implications for academic communities. As we have mentioned, number of citations
is a key determinant of many important decisions in academia because it is generally considered a fair measure of the
impacts and quality of academic papers. Findings from the current research, however, suggest that the papers’ timing of
publication in the year could be a systematic bias that compromises the validity of citation numbers and other related
indices as reliable measures of the quality of academic papers. Our research therefore reveals an overlooked factor that
affects the recognition of scholars’ career attainments. Further, the current mechanism suggests that the citation trap might
hinder scientific advances by making papers published at year-end be less likely to be exposed to the scientific community.
A practical solution to alleviate the influences of the citation trap would be to change the pertinent setting of the time
window options in online academic search engines. Specifically, the search engines should use a “recent n years” option
such that when their users search under a time window, the engines would return papers that were published between the
same date n years ago and the current date. For example, on July 1st, 2019, if a user searches for papers published in the
“recent 5 years”, the engines should return papers published on or after July 1st, 2014, rather than January 1st, 2015. By
this setting, we could at least ensure that papers published at different times in the year have an equal likelihood of being
exposed to the scientific community in online academic search engines.

Funding

This work was supported by the National Natural Science Foundation of China under grant no. 71603046.

Acknowledgment

This paper has benefited from discussions with Peter Aronow and Winston Lin in the course Research Design and Causal
Inference at Yale University. We also would like to thank Xi Chen, Jiacheng Liu and other participants in seminar at Yale
University for helpful comments. James Tierney and Rachel Koh provided excellent assistance.

3
Data from Google Trends are available after 2004. Therefore, we only included data between 2004 and 2010 and the sample size for the IV analysis and
its OLS counterpart was reduced to 69,281.
678 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687

Appendix A

Examples of the time window options in online academic search engines.

C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687 679
680 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687
 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687 681

Table A1
Distribution of the survey
respondents’ location
where the survey was
taken.

Country N

Canada 3
China 83
Colombia 1
France 2
Germany 2
Greece 1
India 1
Ireland 1
Israel 8
Italy 3
Netherland 1
Norway 2
Oman 1
Portugal 1
Singapore 1
Slovakia 2
South Korea 1
Sweden 2
Switzerland 2
Turkey 1
United Kingdom 6
United States 57

Table A2
Distribution of the survey
respondents’ fields of re-
search. A survey respon-
dent could choose more
than one fields.

Field of research N

Economics 82
Marketing 16
Management 20
Accounting 6
Finance 17
Psychology 52
Others 47

Table A3
The proportions of survey respondents who indicated having preferred or avoided month(s) for journal (re-)submissions in different
positions and groups with different number of publications.

Position Doctoral Post- doctoral Assistant Associate Full professor p-value from
student research fellow professor professor chi-square test

Proportion of 4.2% 14.8% 10.6% 11.4% 16.0% 0.487

respondents having
preferred month(s)
Proportion of 6.3% 7.4% 10.6% 2.9% 12.0% 0.640
respondents having
avoided month(s)
N 48 27 47 35 25
Number of 0–2 3–5 6–10 11–20 > 20 p-value from
publications chi-square test
Proportion of 2.3% 8.3% 12.5% 20.6% 10.4% 0.129
respondents having
preferred month(s)
Proportion of 4.5% 12.5% 6.3% 8.8% 8.3% 0.813
respondents having
avoided month(s)
N 44 24 32 34 48
682 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687

Fig. A1. Number of survey respondents who indicated that they would prefer or avoid (re-)submitting their papers to academic journals in each calendar
month.
 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687 683

Appendix B

The mean values of number of pages of the paper, number of authors of the paper, number of references cited in
the paper, and number of letters in the title of the paper for papers published in each calendar month.
684 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687

Appendix C

Results and discussions of the survey

We investigated the prevalence rate of strategic submission by conducting an online survey among scholars. Invitations
to take the survey were sent out in several online academic communities. These include the mailing list managed by the
Society for Judgment and Decision Making, discussion groups consisting of economists from China, and members of the
China Health Policy and Management Society. We had 182 completed responses. The majority of the survey respondents
were from China and the United States (see Table A1). The estimated response rate of the survey is not available because
we do not know the number of people in these communities. About half of our respondents were scholars in economics
and related fields, and the majority of them were from fields of social sciences, including business, psychology, and others
(see Table A2). Thus, the current sample was to some extent representative of authors of the papers in our dataset and also
of scholars of neighboring fields.
In the survey we asked the participants to indicate whether there are certain months in the year in which they either
prefer or avoid (re-)submitting their papers to academic journals. We also asked them to indicate their current position and
the number of papers they had published in academic journals. Detailed instructions and questions of the survey are shown
in Appendix D.
The results showed that the majority of the respondents indicated that there was not any month in the year in which
they preferred (90%; 163/182) or avoided (92%; 168/182) (re-)submitting their papers. The number of respondents who
indicated that they would prefer or avoid (re-)submitting their papers to academic journals in each calendar month was
 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687 685

shown in Fig. A1. Importantly, the proportion of respondents who gave an affirmative answer to the above questions did
not differ across positions or groups with different number of publications (see Table A3). Therefore, the results of our
survey confirmed that strategic submission in terms of timing of publication was neither common among scholars nor was
it associated with their ability or experience.

Appendix D

Instructions and questions of the survey

Dear researchers,
Thank you for taking this short survey. The survey asks a few questions about your practices in journal submissions. The
information you provide in this survey will be confidential and for research use only.
When you submit/resubmit your papers to academic journals, is (are) there any month(s) of the year in which you prefer
to make your submissions/resubmissions?

No, I submit/resubmit my papers as soon as they are ready

Yes, I prefer to submit/resubmit my papers in certain month(s)

(The below question was only presented to respondents who selected “yes” in the above question)
In which month(s) of the year do you prefer to make journal submissions/resubmissions?

January
February
March
April
May
June
July
August
September
October
November
December

When you submit/resubmit your papers to academic journals, is (are) there any month(s) of the year in which you avoid
making your submissions/resubmissions?

No, I submit/resubmit my papers as soon as they are ready

Yes, I avoid submitting/resubmitting my papers in certain month(s)

(The below question was only presented to respondents who selected “yes” in the above question)
In which month(s) of the year do you avoid making journal submissions/resubmissions?

January
February
March
April
May
June
July
August
September
October
November
December

What is your current position?

Doctoral student
Post-doctoral research fellow
Assistant professor
Associate professor
Full professor
686 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687

How many papers have you published in academic journals so far?

Two or below two

Three to five
Six to ten
Eleven to twenty
More than twenty

What is (are) your field(s) of research?

Economics
Marketing
Management
Accounting
Finance
Psychology
Others, please specify

Appendix E

Data from Google Trends for searches for the keyword “spring festival”.

References

Berger, J., 2016. Does presentation order impact choice after delay? Top Cogn Sci 8 (3), 670–684.
Bramoullé, Y., Ductor, L., 2018. Title length. J Econ Behav Organ 150, 311–324.
Card, D., DellaVigna, S., 2014. Page limits on economics articles: evidence from two journals. J Econ Perspect 28 (3), 149–168.
Coupé, T., Ginsburgh, V., Noury, A., 2010. Are leading papers of better quality? evidence from a natural experiment. Oxf Econ Pap 62 (1), 1–11.
Di Vaio, G., Daniel, W., Jacob, W., 2012. Citation success: evidence from economic history journal publications. Explor Econ Hist 49 (1), 92–104.
Diamond Jr., A.M., 1986. What is a citation worth? J Human Resour 21 (2), 200–215.
Ellison, G., 2013. How does the market use citation data? the Hirsch index in economics. Am Econ J Appl Econ 5 (3), 63–90.
Freeman, R.B., Huang, W., 2015. Collaborating with people like me: ethnic co-authorship within the United States. J Labor Econ 33 (3), S289–S318.
Garfield, E., 1999. Journal impact factor: a brief review. Can Med Assoc J 161, 979–980.
Goerg, S.J., Kaiser, J., 2009. Nonparametric testing of distributions—the Epps–Singleton two-sample test using the empirical characteristic function. Stata J 9
(3), 454–465.
Guo, F., Ma, C., Shi, Q., Zong, Q., 2018. Succinct effect or informative effect: the relationship between title length and the number of citations. Scientometrics
116 (3), 1531–1539.
Hamermesh, D.S., Johnson, G.E., Weisbrod, B.A., 1982. Scholarship, citations and salaries: economic rewards in economics. South Econ J 149 (2), 472–481.
Hilmer, J.M., Ransom, R.M., Hilmer, E.C., 2015. Fame and the fortune of academic economists: how the market rewards influential research in economics.
South Econ J 82 (2), 430–452.
Hirsch, J.E., 2005. An index to quantify an individuals scientific research output. Procee Natl Acad Sci 102, 16569–16572.
Huang, W., 2015. Do ABCs get more citations than XYZs? Econ Inq 53 (1), 773–789.
Hudson, J., 2007. Be known by the company you keep cites: quality or chance. Scientometrics 71 (2), 231–238.
 C. Ma, Y. Li and F. Guo et al. / Journal of Economic Behavior and Organization 166 (2019) 667–687 687

King, D.A., 2004. The scientific impact of nations. Nature 430 (6997), 311.
Kousha, K., Thelwall, M., 2007. Google scholar citations and Google Web/URL citations: a multi-discipline exploratory analysis. J Am Soc Inform Sci Technol
58 (7), 1055–1065.
Letchford, A., Moat, H.S., Preis, T., 2015. The advantage of short paper titles. R Soc Open Sci 2 (8), 1–6.
Pinkowitz, L., 2002. Research dissemination and impact: evidence from web site downloads. J Finance 57 (1), 485–499.
Sauer, R.D., 1988. Estimates of the returns to quality and coauthorship in economic academics. J Political Econ 96 (4), 855–866.
Smart, S., Waldfogel, J., 1996. A citation–based test for discrimination at economics and finance journals. National Bureau of Economic Research Working
Paper No. 5460.
Vieira, P.C.C., 2008. An economics journals ranking that takes into account the number of pages and coauthors. Appl Econ 40 (7), 853–861.
Vieira, E.S., Gomes, J.A.N.F., 2010. Citations to scientific articles: its distribution and dependence on the article features. J Informetr 4 (1), 1–13.
Webster, G.D., Jonason, P.K., Schember, T.O., 2009. Hot topics and popular papers in evolutionary psychology: analyses of title words and citation counts in
evolution and human behavior, 1979-2008. Evol Psychol 7 (3), 348–362.