'The Salary Taboo: Privacy Norms and the Diffusion of Information' by Zoe Cullen and Ricardo Perez-Truglia
comments
The diffusion of salary information has important implications for labor markets, such as for wage discrimination policies and collective bargaining. Despite the widespread view that transmission of salary information is imperfect and unequal, there is little direct evidence on the magnitude and sources of these frictions. We conduct a field experiment with 752 employees at a multi billion-dollar corporation to address these questions. We provide evidence of significant frictions in how employees search for and share salary information and suggestive evidence that these frictions are due to privacy norms. We do not find any significant differences in information frictions between female and male employees.
The authors argue
Most employers provide limited information about salaries. Thus, employees’ knowledge
about salaries depends largely on their ability to communicate with each other. However,
there is a widespread belief that the diffusion of salary information is imperfect and unequal.
For example, most employees do not discuss salaries with their coworkers, despite wanting to
be better informed about peers’ salaries (Glassdoor, 2016; PayScale, 2018). These information
frictions are sometimes attributed to firm efforts that discourage employees from discussing
salaries (Gely and Bierman, 2003; Hegewisch et al., 2011). Others argue that the frictions
stem from a “salary taboo”: a social norm around salary privacy that discourages coworkers
from revealing or inquiring about salary information (Trachtman, 1999; Edwards, 2005).
These information frictions are important, because they have implications for a broad
range of labor market phenomena. For example, information frictions can facilitate workplace
discrimination, increase employers’ market power (Danziger and Katz, 1997; Cullen and
Pakzad-Hurson, 2017), and hinder collective bargaining and unionization (Corbett, 2002).
These supposed information frictions also have inspired several policies, such as those that
punish employers when they retaliate against employees who discuss wages with each other
(Pender, 2017; Siniscalco et al., 2017). Despite these important implications, there is little
direct evidence on the diffusion of salary information. We use a field experiment to provide
novel evidence on how individuals search for and share salary information and on the role
that privacy concerns play in these decisions.
Employees can benefit from information about coworkers’ salaries in several scenarios,
such as negotiating salary, switching managers, or searching for new jobs. This information
has a cost, though: employees must spend time and energy to search for it, and they may
face costs for inquiring about sensitive data. We design a novel field experiment to study
these costs and benefits.
To measure employees’ willingness to search for information, we generate an exogenous
shock to the benefits of being informed about salaries by allowing employees to partake
in a game. In the game, employees guess the average salary of a random sample of five
of their peers (e.g., a bank teller guesses the average salary of five other tellers from the
same branch). Employees whose guesses fall within 5% of the true average salary receive a monetary reward. After providing their initial guesses, which they must do immediately,
respondents are offered the opportunity to acquire an extra week to search for information
in the wild and improve their guesses. We elicit the probability of winning the game with
and without the additional week, using self-reported and incentive-compatible methods. The
degree to which employees expect the extra week to increase their probability of winning the
game measures their willingness to search for information in the wild.
We measure the gross benefits from information by eliciting the willingness to pay for an
imperfect but informative signal about the average peer salary, using an incentive-compatible
method. In other words, we provide subjects with the opportunity to acquire readily available
information from the experimenter instead of searching for information in the wild. Last, we
measure the willingness to share information with others. We offer respondents an opportunity
to reveal their own salaries to five peers, and we use an incentive-compatible method to
elicit the willingness to pay to reveal this information (for subjects who prefer to share the
information) or the willingness to pay to conceal this information (for subjects who prefer to
conceal the information).
We cross-randomize two key features of the survey. The first treatment arm allows us
to test differences in the diffusion of salary information, relative to other important career
information. For this, we randomize subjects into two versions of the survey: salary and
seniority. The survey types are identical, except that one asks about the average seniority
of peers instead of the average salary. Just like information about peer salary, information
about peer seniority can be useful to make important career choices, such as whether to ask
for a promotion or search for another job. However, employees may face higher frictions
when searching for and sharing information about salary, compared to seniority, for example,
because of the salary taboo.
The second treatment arm aims to test the rational inattention hypothesis, according to
which individuals search for and acquire new information when they stand to gain from it
(Woodford, 2001; Sims, 2003; Mankiw and Reis, 2002; Reis, 2006). We randomize the size
of the rewards of the guessing game using five different values from $13 to $63 (these and
all other monetary values reported in this paper are expressed in United States dollars using
PPP-adjusted exchange rates from February 2018). This randomization generates exogenous
variation in the benefits of being informed. We test two predictions of the rational inattention
model: higher rewards should increase the willingness to search for information in the wild
and the willingness to acquire readily available information.
We conduct a field experiment with a sample of 752 employees from a large commercial
bank (hereafter referred to as the firm) with thousands of employees, millions of customers,
and billions of dollars in revenues. The firm is typical in some relevant respects. The firm does not have open salary policies and discourages employees from discussing salaries with
each other. Most employees report that they have limited information about salaries and
would prefer the firm to be more transparent. There seems to be a social norm against asking
coworkers about their salaries, and employees rarely discuss salaries with their coworkers.
A number of studies show that these features are common in firms from several countries,
including the United States (Trachtman, 1999; Edwards, 2005; Hegewisch et al., 2011; Glassdoor,
2016; PayScale, 2018).
We find that employees have imperfect information about the salaries of their peers:
the mean absolute error of the guesses reported in the game is 16%. Indeed, this level of
misperception is what we would expect if employees have access only to information about
their own salaries. Although employees are overconfident in their guesses, they are aware that
their accuracy (i.e., the probability of guessing within 5% of the truth) is far from perfect.
We provide evidence that misperceptions are partly due to search costs. When presented
with financial incentives to do so, most individuals are willing to search for information in
the wild. When given an extra week to gather information, the average respondent expects
to increase the probability of winning the guessing game by 23 percentage points. The
evidence reveals that search costs are unequal: when provided with the additional week,
some employees expect to search for information in the wild, but other employees do not
expect to search. And consistent with the rational inattention hypotheses, employees who are
randomly assigned to higher game rewards expect to search more intensively than employees
who are assigned to lower game rewards.
We find that, much like the search costs, the gross benefits of the salary information are
significant and unevenly distributed. The median willingness to pay for the readily available
signal of peer salary is about $13. Consistent with rational inattention, this value is higher
for individuals who are assigned to a higher game reward and thus stand to gain more
from the information. Employees in the bottom half of the distribution, who are willing
to pay less than $13 for the information, seem to be misinformed mainly due to a lack of
interest. On the other hand, the remaining half of subjects highly value the information:
their willingness to pay for the signal has a median of $130 and a mean of $369. These high
valuations suggest that these employees do not search for information in the wild because of
information frictions.
We find that individuals also face significant frictions when sharing information with others.
The willingness to reveal one’s salary to coworkers is both significant and heterogeneous.
Whereas a minority of employees (20%) prefer to share personal salary information with their peers, most (80%) prefer to conceal this information. Moreover, this preference for privacy
can be strong: some employees would reveal their salaries for a small sum of money, but
roughly half would not be persuaded to reveal the information to five peers even for $125.
The preference for privacy is consistent with a salary taboo. Individuals are afraid to ask
coworkers about their salaries, because they understand that most coworkers prefer to keep
their salary information private. Indeed, this interpretation is consistent with our subjective
data. Most respondents report that it is socially unacceptable to ask coworkers about their
salaries and that they feel uncomfortable doing so. Moreover, 89% of respondents believe
that if they ask coworkers about their salaries, they will get asked about their own salaries.
Thus, employees may be afraid to ask coworkers about their salaries because that may force
them to reveal their own salaries, which they dislike.
We find that employees are better informed about peer seniority than about peer salaries.
When guessing salaries, employees are as accurate as they would be if they just reported
their own salaries. This finding indicates that employees do not have access to information
beyond their own salaries. In contrast, when guessing seniority, employees are substantially
more accurate than they would be if they just reported their own seniority. This finding
suggests that employees have access to other information about seniority besides their own
seniority. Moreover, our evidence suggests that, at the margin, employees stand to gain more
from salary information than from seniority information. This evidence suggests that the
difference between salary misperceptions and seniority misperceptions are due to differences
in search costs.
Although it is not the only possible interpretation, the demand for privacy is our favorite
interpretation for the differences between salary misperceptions and seniority misperceptions.
Two main pieces of evidence support this view. First, the revealed-preference evidence indicates
that the topic of salary is substantially more sensitive than that of seniority. The
average employee is willing to reveal personal salary information to a sample of five peers
for $67 and willing to reveal seniority to peers for just $28. Second, the subjective data also
suggest that salary is a more sensitive topic. Whereas 69% of employees find it unacceptable
to ask a coworker about salary, only 6% find it unacceptable to ask about seniority; and
whereas 53% of employees find it uncomfortable to ask about a coworker’s salary, only 5%
find it uncomfortable to ask about seniority.
We find substantial frictions in information diffusion, even though participation in the
game may facilitate this diffusion. For example, the guessing game may provide an excuse to
ask peers about their salaries that mitigates the fears of breaking a social norm or breaking the
company’s disclosure rules. Thus, if anything, our findings may underestimate the magnitude
of information frictions under normal circumstances.
Our last result relates to gender differences in information frictions. This analysis is motivated
by the widespread view that pay secrecy disproportionately affects women (Babcock
and Laschever, 2009) and thus may be one of the factors behind the gender pay gap. Consistent
with this view, survey data indicate that women are less confident than men about
their salary knowledge (Glassdoor, 2016; Cullen and Pakzad-Hurson, 2017). Consistent with
these prior survey findings, our own data indicate that female employees are less confident
than male employees about their ability to guess the salaries of their peers. However, we find
that those differences in confidence do not correspond with any real differences in accuracy.
If anything, female employees are slightly more accurate than their male counterparts. Moreover,
we find that other gender differences are small, statistically insignificant, and precisely
estimated: female and male employees are equally willing to search for information, equally
willing to buy information, and equally willing to share information with peers.
Our study relates to various strands of literature. A large theoretical literature from
economics and management suggests that frictions in the diffusion of salary information can
have important implications for labor markets (Akerlof and Yellen, 1990; Kuhn and Gu,
1998, 1999; Ellingsen and Rosén, 2003; Michelacci and Suarez, 2006; Cullen and PakzadHurson,
2017; Moellers, Normann, and Snyder, 2017). Yet, there is little direct evidence
on the magnitude and sources of information frictions. This study builds on our previous
work documenting significant misperceptions of peer and manager salaries (Cullen and PerezTruglia,
2018).6 This study aims to understand the sources of these misperceptions, with
special emphasis on the role of the salary taboo.
Our study relates to a literature on the diffusion of information in social networks. Several
models explain how individuals form beliefs based on peer-to-peer communication (Bass,
1969; Ellison and Fudenberg, 1995). More recent studies measure social learning in the
field (Mobius and Rosenblat, 2014). Some of these studies artificially create incentives for
information diffusion. For instance, Mobius et al. (2015) recruited college students to play
a “treasure hunt” game in which they earned prizes by collecting information from peers.
Other studies exploit natural incentives for information diffusion. For example, Beaman
et al. (2018) seeded useful information about composting and measured its diffusion in an
agricultural network. These papers show evidence that, even in settings where information is mutually beneficial, its diffusion is highly imperfect. Our contribution to this literature
is twofold. First, we contribute a new method to measure the willingness to search for
information and the willingness to share information with others. Second, we explore the
role of privacy norms for the diffusion of information.
Our paper adds to the literature on the economics of privacy (Acquisti et al., 2016). For
example, Goldfarb and Tucker (2012) show that, even in anonymous internet surveys, some
respondents refuse to reveal information about their incomes and demographics. Athey et al.
(2017) and Adjerid et al. (2013) study the demand for privacy in the crypto-currency market.
They show that even individuals who report that they highly value privacy are willing to give
away sensitive information for small incentives. We contribute to this literature by measuring
preferences for privacy in a context with high stakes (i.e., an employee’s willingness to reveal
personal salary information to coworkers). In contrast to those other contexts, we find a high
willingness to pay for privacy. Perhaps more surprisingly, we find a large heterogeneity in
preferences for privacy, with some individuals willing to pay to reveal their salary to peers
rather than conceal it.
Last, this study relates to literature on wage discrimination. There is a widespread view
that pay secrecy hurts minorities, because it helps employers to discriminate against them
(Phillips, 2009; Colella et al., 2007). This view has led to various efforts to reduce the gender
wage gap, through transparency policies (Colella et al., 2007). However, this argument often
assumes that pay secrecy hinders information access more for women and minorities than for
others. Our evidence does not support this assumption: women and men face similar frictions
and have similar degrees of misperceptions. However, we do find that female employees are
less confident than male employees about the accuracy of their beliefs.
The rest of the paper proceeds as follows. Section 2 presents the conceptual framework.
Section 3 presents the survey design. Section 4 discusses the implementation details. Section
5 presents the results. The last section concludes
