The effect of social gender norms on parental leave uptake intentions: evidence from two survey experiments on prospective fathers and mothers

ABSTRACT We investigate how social gender norms influence parental leave uptake intentions by conducting two separate survey experiments on prospective fathers (N = 877) and mothers (N = 882) in the UK. In a between-subjects design, we manipulate social gender norms by varying information on the average number of days that other fathers and mothers stay at home to take care of a child during the first year after childbirth. We find that when prospective parents (both genders) are exposed to the low staying-home-with-children norm, they plan less parental leave uptake compared to the control (no norm) group. When exposed to the high staying-home-with-children norm, men (but not women) plan more parental leave uptake compared to the control group. We discuss policy implications and suggest directions for future studies.


I. Introduction
Across the world, women take the major responsibility for family and children.Despite a remarkable progress towards gender equality in the labour market during the last decades (Goldin 2014(Goldin , 2021)), women are the ones that usually stay home with small children after childbirth, while men continue to work (Ray, Gornick, and Schmitt 2010;Sundström and Duvander 2002).In the OECD countries (where data is available), fathers make up only approximately 20% of all parental leave recipients. 1 The Nordic countries stand out with the highest share of fathers that take leave after having a child, while almost no fathers take parental leave in some other countries.In the UK, approximately 1/3 of all new fathers take at least some parental leave.
It is well documented that the imbalance between fathers and mothers in responsibility for children contributes to labour market gender inequalities in career opportunities, wages, employment, and retirement benefits (Albrecht, Thoursie, and Vroman 2015;Schönberg and Ludsteck 2007;Rege and Solli 2013).Research shows that the imbalance also affects outcomes for children, who spend relatively less time with their fathers, and the effects tend to persist also in the long term (Huerta et al. 2014;Cools, Fiva, and Kirkebøen 2015).Because of the significant consequences, it is fundamental -not least for policymakers -to understand what factors determine family versus career choices for fathers and mothers.
Early seminal work in economics has pointed to comparative advantages in household work as one major explanation for why mothers specialize in household work and take care of the children, while men specialize in paid work (Mincer and Polachek 1974;Becker 1981).One argument that has been put forward based on Becker's (1981) model of an efficient household is that women make lower human capital investments and, as a result, have a comparative advantage in household work.However, in many contemporary societies, this argument does not conform to reality, since women often are more highly educated than men -a trend that seems to be growing (Goldin 2014(Goldin , 2021)).In addition, women who work full-time in the regular labour market still spend more time than men on household duties (Berardo, Shehan, and Leslie 1987;Gershuny and Sullivan 2003;Sullivan 2000).
Therefore, arguments based on comparative advantages cannot explain why women do most household work.
Other more recently discussed explanations concern social psychological constructs such as gender identity, gender roles, and social norms.Identity is defined as one's sense of belonging to a social category (e.g.men or women; Akerlof and Kranton 2000).Our social identities are associated with specific roles, which according to social role theory incorporate expectations, in the form of descriptive and injunctive social norms (Eagly and Karau 2002).Descriptive social gender norms constitute consensual expectations about what most men and women actually do (e.g.women take most parental leave in a couple), whereas injunctive social gender norms refer to prescriptive expectations about what the different genders ought to do (e.g.women should take most parental leave in a couple; Cialdini, Reno, and Kallgren 1990).Descriptive norms motivate people to act via social information (Cialdini 2007).People may automatically infer that if most people behave in a certain way, this is likely to signal what constitutes an effective and accurate course of action in that setting (Cialdini 2007;Cialdini, Reno, and Kallgren 1990).If women take most parental leave in a couple, while men remain at work, people may infer that this division of labour is effective (e.g. the child is better off when the mother does most of the caretaking).In contrast, injunctive norms motivate people to act via social evaluation (Cialdini 2007).People follow others' behaviour based on their beliefs about what behavior is typically approved and disapproved by others to avoid informal social sanctions (Cialdini, Reno, and Kallgren 1990).For example, parents may decide against letting the father stay at home from work to care for their newborn child because they fear that this would be met by greater social disapproval than if the mother would stay at home.
We argue that both types of social norms should matter for parental leave uptake.Parents look at other parents' parental leave uptake behaviour to obtain information on what division of parental leave days between themselves and their partner would be good for their child, but also to determine the social appropriateness of a particular division.
The current research investigates how social gender norms influence family versus career choices by conducting two separate survey experiments on prospective fathers (N = 877) and mothers (N = 882).We study both prospective fathers and mothers since the likelihood of success of any policy reform with the goal of achieving gender equality will depend on both fathers' and mothers' behaviour.We conduct separate experiments on prospective fathers and mothers because the prevailing social gender norms for them are different, as we will discuss below.Participants in the UK aged 25-35 were recruited using the Prolific platform.In the two experiments, we randomly manipulated the social gender norm by varying the information provided to our research participants.Specifically, the manipulation concerned how many days fathers and mothers, respectively, in the UK stay at home to take care of a child during the first year after childbirth.In the experiments, one-third of the sample was presented statistics, i.e. the social norm treatment, about the parental leave usage among fathers (or mothers) which was below the average perception of how many days they use, whereas another third was presented statistics depicting a parental leave usage above the average.We also included a control group, which received no social norm information.After the participants were randomly assigned to one of the three conditions, they indicated how many days they would stay at home taking care of a child during the first year after childbirth, if they would become a father or mother.This design allows us to estimate how signalling social norms influence parental leave uptake intentions, compared to when no social norm is explicitly signalled.Moreover, it allows us to investigate whether parental leave uptake intentions can be affected both downwards and upwards (i.e. by the low and high social norm manipulations).
Indirect evidence supports the idea that genderrelated social norms may influence labour market outcomes and family relationships.Fernández (2013) argues that the dynamics of women's labor supply and social norms in the US are connected historically.Using US data, research by Bertrand, Kamenica, and Pan (2015) suggests that when injunctive norms are violated, this may have negative consequences for women's labor supply.For example, women are less likely to be in the labor force when their potential income is higher than their husband's.This research also shows that marriage rates tend to decline when women earn more than men.For Sweden, Folke and Rickne (2020) document that divorce rates are higher for women (versus men) following promotions.For the US, divorce rates have been higher for couples where the woman earns more than the man (Bertrand, Kamenica, and Pan 2015), although this does not appear to be the case for couples who have married during recent decades (Schwartz and Gonalons-Pons 2016).
More closely related to the current research, there are quasi-experimental studies suggesting that social gender norms among co-workers could matter for parental leave usage.These studies use policy reforms as natural experiments, or other instruments, to address the challenge that norms are unlikely to be exogenous in observational studies on parental leave usage. 2 Dahl, Løken, and  Mogstad (2014) use the Norwegian 1993 reform on paid parental leave, which arguably creates exogenous variation in social gender norms regarding parental leave usage among co-workers.Their findings show that the likelihood that a father takes parental leave is affected by the share of co-workers that take parental leave.Welteke and Wrohlich (2019) investigate social gender norms among female coworkers in Germany.For identification, they use a parental leave policy reform in 2007.Their findings are that mothers use additional parental leave days when other mothers at the workplace increase their usage.Carlsson and Reshid (2022) use an instrument variable constructed from partially overlapping networks -workers at the same firm and individuals in the same familyto investigate coworker peer effects on parental leave usage in Sweden.They find that both mothers and fathers use more parental leave days when coworkers of the same gender increase their usage.
Another strand of literature of high relevance to our study consists of experimental studies (mainly in social psychology) that have directly manipulated social norms to assess causality.There is ample experimental research evidence showing that social norms influence judgements and behaviours in numerous important life domains.For example, Goldstein, Cialdini, and Griskevicius (2008) demonstrate that conveying social norms can be a powerful tool for increasing environmentally friendly behavior.By changing towel reuse signs in hotel rooms, the study shows that towel reuse behavior increases substantially when social norms are manipulated by stating that '75% of hotel guests have reused their towels' compared to when the towel reuse signs contain industrystandard pro-environmental appeals (e.g.'Help save the environment').Social norms can also have an impact on charitable giving.Agerström et al. (2016) show that monetary donations to nonprofit organizations increase when people learn that most others who have been solicited for a donation have also contributed to the charity.The norm effect increases further the more similar previous donors are to the person solicited for a donation.Voting behavior is another area where it has been shown that social norms matter.Gerber and Rogers (2009) provide field experimental evidence showing that people's intention to vote in an upcoming election is influenced by their perception of whether others are going to vote.Prospective voters who are exposed to norm information conveying that voter turnout in the upcoming election is likely to be high declare higher voting intentions than prospective voters exposed to norm information conveying that voter turnout is likely to be low.The results further suggest that the norm effect mainly applies to infrequent and occasional voters.
Based on social norm theories (e.g.Cialdini, Reno, and Kallgren 1990) and a large bulk of experimental research showing that social norms play an important role in driving people's behaviour across numerous other domains, our two preregistered main hypotheses are: 3 H1: The planned number of days staying at home taking care of the child during the first year after childbirth should be higher for those treated with the high staying-home-with-children norm compared to those in the control group.

H2:
The planned number of days staying at home taking care of the child during the first year after childbirth should be lower for those treated with the low staying-home-with-children norm compared to those in the control group.
To our knowledge, our study is the first that experimentally manipulates social gender norms to study their effect in the context of parental leave.Although previous quasi-experimental research suggests that norms could matter for parental leave uptake, an experimental approach offers a higher degree of control over confounding variables and should therefore be superior in establishing a causal effect of norms.Experiments are therefore an important complement to quasiexperimental research which has the strength that it is based on data on real parental leave usage.If the results converge, we can be more certain that social norms matter for parental leave uptake.
We continue by presenting the experimental approach (Section II) followed by the results of the experiment on prospective fathers in Section III and the experiment on prospective mothers in Section IV.In Section V, we discuss the results of the two experiments in terms of their broader implications for the gender imbalance in responsibility for children, gender inequalities in the labour market, and policy.Finally, we conclude, discuss some limitations, and provide directions for future research (Section VI).

II. Method
We conducted two separate survey experimentsone on males (prospective fathers) and one on females (prospective mothers) -to investigate how the number of days they would stay at home taking care of the child if they would become a parent is affected by social gender norms.The two experiments had the same overall design, although they differed considering sampling and treatment.Each experiment had its own research plan, and these were preregistered at the American Economic Association's registry for randomized controlled trials (AEARCTR-0006692 and AEARCTR-0006838) before collecting any data.The preregistered research plans describe the experiments and contain research design, hypotheses, variable operationalizations, model specifications, robustness checks, and handling of missing data.Before turning to the experimental design, we give a contextual background by describing the parental leave system in the UK.

Parental leave in the UK
Most workers in the UK are entitled to statutory parental leave. 4To qualify for statutory parental leave, a worker must have an employment contract and give the employer a correct notice (at least 15 weeks before having a child).To qualify, the worker must also have earnings of at least £123 per week and been working for the last 26 weeks.A worker's employment rights (such as pay raises or return to work) are protected during the statutory parental leave.
Mothers who qualify for statutory parental leave are eligible for 52 weeks of leave.A mother must use at least 2 weeks of statutory parental leave right after birth, but can then decide how much to use of the remaining 50 weeks.
While fathers are only entitled to 2 weeks of statutory parental leave, it is possible to share leave days between the parents.If a mother wants to share statutory parental leave with the father, the mother has to give away leave days to the father and use less than the 52 weeks.It is possible to give away up to 50 weeks of statutory parental leave.Mothers and fathers can take leave in blocks separated by periods of work, or take their leave in a sequence.
In an international comparison, the statutory parental leave for both mothers and fathers in the UK is substantially lower than the OECD average. 5oncerning actual parental leave usage in the UK, there are no official statistics available, since parental leave usage is administrated by the employers. 6It is also the employers who pay the statutory pay to employees.While the pay is 90% of the average weekly earnings (before tax) for the first six weeks, it is typically much lower for the remaining period of leave.7

Design
Participants in the experiments were first exposed to the norm manipulation and then asked to report parental leave intentions.For males, the norm manipulation concerned how many days on average other fathers stay at home taking care of the child during the first year after childbirth, while for females it concerned other mothers.Using a betweensubjects design, the participants in both experiments were randomly assigned to one of three groups.The first constitutes a control group for which no explicit information about how many days other parents in the UK stay at home taking care of the child during the first year after childbirth was presented.Males (females) in treatment group 1 learned that the number of days other fathers (mothers) stayed home was high.Males (females) in treatment group 2 learned that the number of days other fathers (mothers) stayed home was low.
We conducted two pre-studies (N = 150 in each) measuring the perceived descriptive social norm regarding fathers' and mothers' parental leave uptake which targeted the same populations as in the real experiments.In the first pre-study, we asked male participants the following question: How many days do you think fathers in the UK on average stay at home to take care of their child during the first year (365 days) after childbirth?On average, the respondents estimated that fathers stay at home for approximately 40 days.In the second pre-study, we asked female participants the following question: How many days do you think mothers in the UK on average stay at home to take care of their child during the first year (365 days) after childbirth?In this case, the average respondent answered that mothers stay at home for approximately 236 days.
We used the results from the two pre-studies to create a low and high treatment norm for each experiment.Here we assumed that participants in the control groups are likely to consult some norm reference value based on their own assumptions about how many days fathers or mothers stay home on average.This norm reference value should be rather similar to the perceived norm value uncovered in the pre-studies on the same populations.In both the experiments on males and females, we then set the high (low) stayinghome-with-children norm treatment such that the number of days fathers or mothers stay home should be substantially higher (lower) than participants' expectations in the baseline condition.In the experiment on males and the low treatment norm, we used a number which is both substantially lower than 40 (the mean in the pre-study on males) and above zero, which would be unrealistically low.We decided to set the low treatment norm to 10 days.To obtain symmetry around the mean we set the high treatment norm to 70 days.This corresponds to a distance of ±2/3 of a standard deviation from the mean.Using the same distance from the mean (in terms of standard deviations) in the experiment on females, we set the low and high treatment norms to 180 and 290 days in this case.
The material which was presented to fathers who received the low norm treatment looked as follows: Consider a couple consisting of a man and a woman who are expecting their first child.After birth the mother and/or father usually take time out from work to stay at home for a period to take care of the child.[Fathers in the UK on average stay at home 10 days to take care of the child during the first year (365 days) after childbirth].For males in the high treatment group the only difference was that 10 days was replaced by 70 days, while for males in the control group the sentence in brackets was omitted.We used the corresponding material for the experiment on prospective mothers, the only differences being the norm reference group (mothers) and the number of days (180 versus 290).
While our social norm manipulation follows the format of descriptive norms (i.e.what most people do), it may also to some degree contain information on what behaviour is accepted by society.After all, if many people behave in a certain way, people may to some extent infer that this behavior is socially permissible (Cialdini 2007).Note that our experiments are not designed to disentangle the effect of descriptive norms from that of injunctive norms, but rather to assess a more general effect of social norms on parental leave uptake intentions.
After being presented with the norm manipulation in the experiments, participants completed the outcome measure by answering the following question: If you become a father/mother, how many days (0-365) would you stay at home taking care of the child during the first year after childbirth?
The questionnaire also contained a set of questions about background characteristics of the participants and, finally, a manipulation check. 8Those who failed the manipulation check or completed the survey within 30 seconds were discarded from the sample and analyses (as described in the preregistered research plan).The materials used in the survey experiments including all survey questions are found in appendix Table A1.

Population, sampling, and procedure
The targeted population in the experiment on prospective fathers were male UK citizens aged 25-35.The only difference in the experiment on prospective mothers was that it instead targeted females.In both experiments, we randomly sampled 1,000 participants 9 from the Prolific platform (www.prolific.co).The survey experiment was then conducted through an online survey using the Qualtrics survey software.All data were provided anonymously.Participants were introduced to a study on parental leave.If consenting to participate, participants envisioned the scenario described above.A debriefing concluded the session.Data were collected in November and December 2020.Sample summary statistics are presented below.

Statistical analysis
We estimate treatment effects in a regression framework using ordinary least squares.To test H1 and H2, we estimate the following regression on the planned number of days staying at home taking care of the child during the first year after childbirth: Here, y i is the outcome for individual i, NORM H i is a dummy variable equal to 1 if treated by the high staying-home-with-children norm (else 0), NORM L i is a dummy variable equal to 1 if treated by the low staying-home-with-children norm (else 0), and ε i is the error term which is assumed to have mean zero and be independent of the treatment variables.Hypothesis H1 implies β H > 0 and hypothesis H2 implies β L � 0.

Balance tests and summary statistics
Following the preregistered research plan, we start by conducting balance tests across a set of potentially important predictors of the outcome variables.Of course, we have no reason to believe that the inbuilt randomization tool in the Qualtrics software that we used would have failed to randomize the participants to the low and high treatment conditions.We examine balance across age, and employment status (indicators for full-time employed, part-time employed, unemployed, and other employment status).We conduct two balance tests.In the first, we regress an indicator of whether the participant was assigned to the high 8 Subjects in the treatment groups passed the manipulation check if they correctly answered the following question: How many parental leave days do fathers (mothers) in the UK take on average in the first year of their child?. 9 We calculated the statistical power of detecting H1 and H2 (the main hypotheses) simultaneously as outlined at the EGAP webpage (http://egap.org/content/power-analysis-simulations-r) using the results from the pre-studies.The assumptions and the other details of the power calculations are documented in the preregistered research plans.We concluded that we needed approximately 900 respondents in the experiment on males to have 80% power to detect that two main treatment effects simultaneously.In the experiment on females, 450 respondents were needed.In the end, we decided to sample 1,000 respondents in each experiment.
treatment (as opposed to being assigned to the low treatment or being in the control group) on the covariates.In the second, we instead regress an indicator of whether the participant was assigned to the low treatment (as opposed to being assigned to the high treatment or being in the control group) on the covariates.
The results of the two balance tests are reported in Columns 1 and 2 of Table 1.None of the tests show any strong support for an imbalance.Although a few individual parameters are weakly significant, F-tests of joint significance do not reject the null that all parameters are equal to zero.We conclude that the randomization appears to have worked as intended.
Table 2 presents summary statistics for the experiment on males.In the end, 877 subjects were included in the analysis out of the initial sample of 1,000.Following the preregistered research plan, 123 participants were excluded because they did not finish the survey within the set timeframe or did not pass the manipulation check at the end of the survey.The mean of the reported planned number of days to stay at home with a child during the first year after childbirth for males was approximately 97 days.From appendix Figure A1, which shows the distribution of this variable, it is clear that this variable's distribution is skewed to the right.The majority of males plan to stay home less than 100 days, although there also are males that report they would stay home for 365 days.
As expected, approximately one-third of the participants are found in the control, high

Main results
As noticed above, the dependent variable is skewed to the right for males (appendix Figure A1).Therefore, following the preregistered research, we transform the dependent variable by taking the natural logarithm before we estimate Equation (1).However, to facilitate a comparison of the treatment effects with the second experiment on females and show that the results are robust, we also estimate Equation ( 1) using the non-transformed variable and using a dummy variable around the median of the planned number of days as the dependent variable.Table 3 presents the main results of the experiment on males, i.e. the result of testing hypotheses H1 and H2 for the high and low norm treatment effect.In Column 1, we see that the point estimates for respondents in the high and low treatment groups equal approximately 0.24 and −0.64 (both are significant at the 1%-level).The interpretation of the coefficients is that males treated by the high treatment norm report that they plan to stay home approximately 24% more days with a child compared to those in the control group, while males treated by the low treatment norm plan to stay home approximately 64% fewer days.We obtain qualitatively the same results if we instead use the untransformed variable or the dummy indicator as the dependent variable.The constant in Column 2 shows that males in the control group report that they would stay home for approximately 106 days on average.The treatment effects reveal that those in the high treatment group plan approximately 11 days more (not significant) and those in the low treatment group plan for 39 days less (significant at the 1%-level).The constant in Column 3 shows that 52% of those in the control group plan for more than the median number of days, while this figure is 22% points higher in the high treatment group and 28% points lower in the low treatment group and (both differences are significant at the 1%level).Note that R-square is highest in Column 1 and 3.In conclusion, for males, we find support for both H1 and H2.

Balance tests and summary statistics
The preregistered research plan for the experiment on females corresponds to the one for males.Thus, we again start by conducting balance tests across the same predictors of the outcome variables as for males.As before we conduct one balance test in which we regress the indicator of whether the participant was assigned to the high treatment group (versus the control or low treatment group) on the covariates and one in which we instead use the low treatment indicator as the outcome.
The results of the balance tests for the experiment on females are reported in Columns 1 and 2 in Table 4. None of the tests show any signs of imbalance; no parameters for individual covariates are significant and not the F-tests of joint significance either.We conclude that the assignment of the high and low treatment conditions appears to be random, as intended.
Summary statistics for the experiment on females are shown in Table 5.We sampled 1,000 individuals in this experiment as well.However, 118 of them did not complete the survey experiment within the set timeframe or did not pass the manipulation check at the end of the survey.After excluding these individuals (following the preregistered research plan), we are left with 882 female subjects.As expected, the mean of the reported planned number of days to stay at home with a child during the first year after childbirth − 297 days with a standard deviation of 85 days -is much higher in the experiment on females compared to in the experiment on males.The shape of the distribution of the planned number of days (see appendix Figure A2) differs as well.A very small number of participants in the female experiment report fewer than 180 days (i.e. six months).Approximately half of the participants report that they would stay home 365 days (i.e. the whole year), while the other half report a number less than 365 days.As expected, the control, high treatment, and low treatment groups are of similar size.Although the majority (56%) of the participants  report that they are full-time employed, the share of part-time employed is larger than in the experiment on males (21% compared to 9%).Finally, 71% of the participants report that they have a college education, 15% that they are students, and 40% that they have a child.

Main results
Since approximately 50% of the participants in the experiment on females report that they would stay at home the whole year, i.e. 365 days, it is natural to code the dependent variable as a dichotomous variable, which takes the value of 1 if the subjected reported 365 days and otherwise 0. However, we did not foresee this distribution and hence did not register a transformation of the outcome into a dummy variable in the preregistered research plan.Therefore, we present results both for the continuous variable (in the research plan) and this dummy variable.It turns out that the values of this dummy variable and of a dummy variable around the median number of planned days are identical, i.e. it is the same variable.
As in the experiment on males, we estimate regressions corresponding to Equation (1) to investigate hypotheses H1 and H2.The results are presented in Table 6.The constant in Column 1 shows that females in the control group on average plan to stay home for 301 days.From this column we also see that there is no evidence of that being treated with the high treatment norm has an effect on the number of reported days -the high treatment coefficient is small in magnitude and insignificant.However, the low treatment norm has a negative effect (significant at the 5%-level) on the planned number of days.The point estimate is approximately −16, which means that females in the low treatment group report that they plan to stay home approximately 16 days fewer compared to those in the control group.
When we use the dummy coding around the median as the dependent variable the results are qualitatively similar.For those in the low treatment group the fraction with planned days below the median is 9% points lower than in the control group (significant at the 5%-level), while the high treatment coefficient is insignificant.Note that the R-squares are substantially lower in this experiment than in the experiment on males.To sum up, in the experiment on females, we find support for H2 but not for H1.

V. Discussion
Across the world, mothers take the major responsibility for family and children and it is well documented that this contributes to labour market gender inequalities in career opportunities.The effect is not limited to parents but seems to affect the outcomes of children too.The current research examined parental leave uptake -one domain in which gender differences are substantial.It appears to be the first experimental research showing that social norms influence parental leave uptake intentions for both prospective fathers and mothers.On a general level, the results align with previous quasi-experimental research aiming to assess the impact of social gender norms, or peer effects, on actual parental leave uptake (i.e.Dahl, Løken, and Mogstad 2014;Welteke and Wrohlich 2019;Carlsson and Reshid 2022).
While the current experiments do not measure actual parental leave uptake, they should present compelling evidence concerning the causal pathway from social norms to parental leave uptake intentions.Interestingly, we find that both prospective fathers and mothers react to social gender norms.In the experiment on prospective fathers, the results support both hypotheses H1 -that the planned number of days staying at home taking care of the child is higher for those treated with the high staying-home-withchildren norm -and H2 -the corresponding hypothesis for the low staying-home-withchildren norm.In the experiment on prospective mothers, the results reveal a picture that is somewhat different.While hypothesis H2 is supported, there is no support for H1, since the estimated high treatment effect is both small in magnitude and insignificant.
An obvious question is why females do not react to the high treatment norm.We notice that in this experiment approximately 50% of the participants report that they plan to stay home the whole first year after childbirth if they would get a child.Thus, for these women, it is not possible to increase the number of days further since they already are at the maximum.This ceiling effect could potentially conceal the impact of the high treatment norm among females.
From a policy perspective, our results suggest that social gender norms could be a potentially strong force in determining how much parental leave fathers and mothers take.If the policy goal is to create a more equal usage of parental leave between fathers and mothers, manipulating social gender norms could be a viable and effective tool.The finding that fathers' parental leave uptake intentions increased when exposed to the high social norm seems promising from an equality perspective.Relatedly, the finding that mothers' parental leave uptake intentions decreased when exposed to the low social norm suggests that social norms signalling a more modest parental leave uptake among mothers could be another, complementary path towards equality.A practical implementation of this could be achieved through information campaigns that highlight actual groups that already take up a desirable number of parental leave days.In doing so, the most natural way of signaling norms would be in the format of descriptive norms (what other fathers actually do) rather than injunctive norms (what society thinks they should do).Besides being less prescriptive and provocative, descriptive norms should be easier to back up with objective data.For example, in many countries, actual parental leave uptake data for desirable groups would be readily available from administrative registries, whereas social approval data would not.In addition, what other parents actually do could also to some extent send a signal regarding what is socially appropriate, and thus indirectly capitalize on injunctive norms too.
Our survey experiments come with a few limitations.One potential limitation is that, although we do not rely on convenience samples (e.g.university students), we cannot rule out self-selection into the study, which would limit the external validity of the results.Yet, it is unclear how self-selection would influence the results in a systematic manner.One possibility is that mainly people who find parental leave important accept to participate in the study.If so, this could have contributed to an underestimation of a social norm effect, given that parental leave intentions would already be high when enrolling in the experiment.
Second, while we have argued that the practical usefulness of signalling descriptive, as opposed to injunctive, social norms should be greater, future research needs to test the validity of this argument.One limitation of the current findings is that they are silent with respect to the relative effectiveness of descriptive versus injunctive norms.Future experimental research should manipulate and test the effectiveness of injunctive vis-à-vis descriptive norms, as well as their combined effect, on parental leave uptake intentions.Such research may also want to study other potential moderating variables that constitute proxies of gender stereotypes and values.Culture (independent vs. interdependent) and political affiliation (liberal vs. conservative) may be significant moderators since they differ in the extent to which they encourage women (vs.men) to be nurturing, caring, and concerned for others.Identification of theoretically driven moderators could provide important knowledge about when injunctive norms could exert an important role.Additionally, future studies should examine which type of social norm tends to be most salient among parents and how such perceptions drive parental leave behaviour.
Third, the potential of social norms to influence parental leave uptake should be limited by different constraints which parents face.Employment status or substantial differences in earnings between parents would be examples of major constraining factors.In such cases, social norms would likely have little impact (Cialdini, Reno, and Kallgren 1990).Arguably, there should, however, be many cases when parents are not too constrained for norms to have a practically important societal impact, particularly in societies with lower gender income disparities.The fact that our experiments were conducted on UK participants instead of participants from, e.g. the Nordic countries, may have put greater restrictions on the social norm effect.Future research may want to examine whether the social norm effect on parental leave uptake varies across countries and cultures.Relatedly, there is a need to understand at what stage economic differences within couples become too large for norms to have an effect on parental leave uptake.
A fourth limitation is that we only measured behavioural intentions rather than actual behavior.That is, we asked the participants about how many days they plan to stay home with a child if they would become a parent, which may not correspond to the actual days they would take up in the future, which is unknown to us.Hence, field experimental research is needed.Researchers could, for example, manipulate descriptive social gender norms (in step 1) and then measure actual parental leave usage among fathers and mothers (in step 2).

VI. Conclusion
We conclude by noticing that the results of the current experiments with presumed clean identification of causal relationships are largely consistent with the results of previous quasiexperimental studies measuring actual parental leave uptake, providing additional support for the idea that social norms matter for parental leave uptake.Thus, when prospective parents contemplate how many days they will stay at home with their child in the future, they seem to look at other parents' behaviour for guidance.This result has potentially important policy implications.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Funding
Financial support from The Swedish Research Council (grant dnr 2018-03487) is gratefully acknowledged.

A2. Secondary hypothesis H3: career-versus familyoriented personal values
In the preregistered research plan, we also state a (secondary) hypothesis H3.This hypothesis is motivated by the fact that there is limited research investigating whether norm effects can be moderated by personal values (but see Lönnqvist, Walkowitz, Wichardt, Lindeman, and Verkasalo 2009).We argue that the relevance of information about other parents' parental leave usage could depend on personal values and hence subjects may react differently to norm information.
In our context, we argue that those with careeroriented (versus family-oriented) personal values could react more strongly to the norm manipulations.First, in the presence of uncertainty about what the effect of absence may be on the future career, information obtained about other parents' parental leave behaviour may be particularly important for those with careeroriented personal values.Second, those with careeroriented personal values tend to work in jobs which require specific skills where it usually is more costly to find a replacement for a worker on parental leave.Thus, there could be stronger reactions in these jobs from employers and co-workers to workers who take parental leave for longer periods.As a result, how much parental leave other parents take may be a particularly important source of information for parents with career-oriented personal values.Third, it is possible that those with family-oriented personal values to a larger extent already have planned for a situation with children and a family, how much parental leave to take, and how to divide responsibilities within the couple.These parents may put lower weight on information on other parents' parental leave behavior.Since we do not have a direct measure of personal values, we use level of education, which we observe, as a proxy for career-versus family-oriented personal values.Our third (secondary) hypothesis is: H3: The social norm effect should be stronger for highly educated individuals.
To test H3, we estimate the following regression in which COLLEGE YES i is an indicator equal to one if the participant has a college education and otherwise equal to zero, and COLLEGE NO i is the negation of COLLEGE YES i .For pedagogical reasons and to ease the interpretation of the interaction effects, we do not include the main variables themselves (NORM H i and NORM L i ) in Equation ( 2).In this setup, the coefficients of the first two interaction variables of Equation (2) measure the high norm treatment effect among those without and with a college degree, while the coefficients of the next two interaction variables measure the low norm treatment effect among those without and with a college degree.This setup makes the coefficients in Equation (2) directly comparable with the coefficients in Equation (1) (reported in Table 3 and Table 6

A2.1 Results for H3 in the experiment on males
To test hypothesis H3 we estimate the regression corresponding to and the results are presented in appendix Table A2.For the high treatment group there is no evidence of differences in the effect between those with and without a college education (compare the coefficients in Row 1 and 2; the p-value for the null that the coefficients are equal is .247).In contrast, for the low treatment there is some evidence of that the negative effect is stronger among those with a college education (compare Row 3 and 4).In terms of their absolute values, the low treatment coefficients are substantially larger among those with college education compared to those without.These differences are statistically significant in Column 2 and 3, but not in Column 1 (see p-values in the table).

A2.2 Results for H3 in the experiment on females
As for males, we estimate the regression corresponding to to test hypothesis H3 for females.The first two rows in Column 1 of appendix Table A3 show that for those that received the high treatment norm there is no evidence of a difference between those with and without a college degree.In contrast, the negative effect of receiving the low treatment norm is much stronger among those without a college education compared to those with a college education.The difference in the low treatment effect between the two education groups is statistically significant (p = .036).In Column 2, where the dummy around the median of planned days is used as the dependent variable, the pattern is very similar.We conclude that there is no support for hypothesis H3; for the low treatment group the pattern is the opposite of what H3 predicts and for the high treatment group there is no evidence of that the education level matters in any direction.

A2.3 Interpreting the results for H3 for males and females
Taken together the results do not support hypothesis H3.However, there are still differences between individuals with different education levels.Although admittedly speculative, we have tried to interpret these differences.
In the experiment on males, we find that the low treatment norm effect is stronger for highly educated individuals.A possible explanation is that males with career-oriented personal values prefer to stay home as few days as possible but experience a cost with decreasing the number of days more because that would not be accepted by their surroundings, i.e. there is a limit regarding the minimum number of days that is acceptable.However, when they see that other fathers' uptake is low, they decrease the number of days they plan to stay home with a child.Possibly, they use this norm information as an opportunity to maximize the investments in their own careers.
In contrast, in the experiment on females, the low treatment norm effect appears concentrated to the group without a college education.It is likely that individuals in the surroundings of these participants on average lack college education and have more conservative family values where there is a strong norm that women take the bulk of the parental leave.As a result, these participants could expect a substantial cost of violating the norm by staying home for too few days.However, when they observe that other mothers take a lower number of days, they also lower the days they plan to stay home with a child because the expected cost of doing so decreases.

Table 1 .
The probability of receiving low and high treatment.Experiment on males.The regressions are estimated using ordinary least squares.In Column 1, the dependent variable is an indicator equal to 1 if the respondent was assigned with the high treatment (as opposed to being assigned the low treatment or being in the control group).Similarly, in Column 2, the dependent variable is an indicator equal to 1 if the respondent was assigned with the low treatment.The regressions include no other covariates than those listed in the table.Robust standard errors are reported in parentheses.

Table 3 .
Planned days to stay home with a child.Experiment on males.
six missing values on the dependent variable (six respondents reported zero planned days for which the natural logarithm is undefined).The regressions are estimated using ordinary least squares.The regressions include no other covariate than the dummy indicators for low and high treatment.Robust standard errors are reported in parentheses.*** p<0.01, ** p<0.05, * p<0.1.

Table 4 .
The probability of receiving low and high treatment.Experiment on females.See also the notes below Table1.Robust standard errors are reported in parentheses.

Table 6 .
Planned days to stay home with child.Experiment on females.See also the notes below Table3.Robust standard errors are reported in parentheses.
).For example, β H in Equation (1) is the high norm treatment effect in the full Planned days to stay home with a child.Experiment on females.YES > β H;NO and β L;YES < β L;NO .Note that the expectations on the high and low norm treatment effects are symmetric, since β H is assumed to be positive and β L to be negative (see the motivation of of H1 and H2 and Equation 1 in the paper).

Table A2 .
Planned days home with child by education.Experiment on males.The regressions include no other covariates than those listed.See also the notes below Table 3. Robust standard errors are reported in parentheses.*** p<0.01, ** p<0.05, * p<0.1..

Table A3 .
Planned days home with child by education.Experiment on females.See also the notes below Table4.Robust standard errors are reported in parentheses.