Perceived returns to job search ☆

This paper provides new evidence on workers’ perceptions of the returns to job search eﬀort using hypothetical vignettes. This allows us to overcome limitations with survey data on realized behavior in which search eﬀort and reservation wages may be endogenous to perceived job ﬁnding rates. The perceived job ﬁnding probability is nearly linear in hours searched and only slightly concave for most respondents. While workers are over-optimistic about the probability of receiving a job oﬀer conditional on any search, they perceive the marginal return to additional search hours as positive but comparably low. Job seekers receiving an oﬀer, update their perceived returns upwards, while the beliefs of unsuccessful searchers regress towards the direction of the mean. We ﬁnd little evidence that novel aspects of the pandemic recession have fundamentally changed workers’ motivations for job search: that an existing job is expected to end or has unsatisfactory pay are the primary motives for on-the-job search. On the contrary, workers’ ability to do their tasks from home is not a strong predictor of job search nor a signiﬁcant motive for switching occupations.


Introduction
In order to find a new job, one typically has to search for a job.This takes time and effort.In most search-and-matching models, job seekers determine their search intensity to balance the expected benefits and costs of search ( [1-3] ).There are many dimensions to job search beyond simply the number of job applications sent.Which occupations to target, at which wage rate, and whether to invest effort now or later, are all relevant dimensions to determine the speed of labor reallocation and new patterns of labor market sorting.Understanding workers' perceived returns to job search is thus crucial in general, but also specifically relevant for understanding the labor supply response to sectoral reallocation and economic recovery in the post-Covid economy ( 4,5 ).
In this paper, we document workers' search behavior and their perceptions of the returns to search.To do so, we collect a two-wave panel dataset of real-time survey evidence on search behavior for workers searching on the job and for unemployed workers in the UK.At the time of data collection, the UK had emerged from formal stay-at-home orders and non-essential retail, restaurants and pubs had reopened.At the time of our first survey, 12% of employees were on furlough compared to a high of 32% in June 2020.The Government was running active advertising campaigns encouraging workers to switch into expanding areas of the economy. 1he Covid-19 pandemic has caused a significant change to the level and pattern of labor demand (6) .Many advanced economies have simultaneously experienced a rise in unemployment alongside labor shortages in key sectors of the economy.As public health experts increasingly predict that the virus will become endemic (7) , labor market policy has shifted from preserving pre-pandemic employer-employee matches to encouraging displaced workers, and workers whose jobs have become durably unproductive, to find new jobs in resilient sectors of the economy.
The pandemic has also significantly affected labor supply.Social distancing restrictions have seen more workers work more from home.Early polling suggests that many want to continue working from home, at least part time ( 8,9 ).Covid has also made new dimensions of health inequality salient, potentially with consequences for the preferences of workers over the characteristics of jobs they value (10) .Frictions in adjusting working patterns within a job may thus have created new incentives for workers to switch into new jobs that better align with their preferences.
Our goals are twofold.First, we provide evidence on workers' perceptions of the returns to search.To elicit perceived returns to job search, we use hypothetical scenarios that allow us to estimate individual perceived job-search functions, i.e. how likely individuals are to receive a job offer depending on the time searched and reservation wage.This allows us to overcome limitations with survey data on realised behaviour in which search effort and reservation wages may be endogenous to perceived job finding rates (11) .More specifically, we can approximate (i) the perceived level by asking how likely they will receive a job offer depending on a given number of hours searched, (ii) the shape of the function by varying the number of hypothetical hours searched within respondents, (iii) how the level and shape depend on the wage by varying target wages in the scenarios, and (iv) the interaction with the state of the economy by asking about search now versus later and the expected state of the economy in the future.As our survey design includes a longitudinal dimension, we can assess whether a worker's search was successful and the degree to which this correlates with prior perceptions.Second, we analyze the degree to which the novel aspects of the pandemic recession and health shock have altered workers' search behavior and their preferences over jobs.
Concerning individual perceived returns to search, we find that respondents believe they have a 32% chance of finding a job in one month if they search for 5 hours per week.The perceived job finding probability is nearly linear in hours searched and only slightly concave for most respondents.However, the marginal return to additional hours searched is comparably low: on average, there is only a 14 percentage points (p.p.) increase in job finding probabilities for increasing search effort from 5 to 15 hours per week, i.e. for each of the additional 10 hours spent searching, the perceived probability of finding a job only increases by 1.4 p.p.This is consistent with [11] who document that only 13% of job seekers perceive a positive marginal return to search.The fact that the perceived returns to the extensive margin of searching are high but the marginal return to additional hours is comparably low might explain why timeuse surveys tend to reveal low search intensities amongst job seekers.On the one hand, our findings show that, absent more detailed information, it is defensible to engage in the common practice in structural search-and-matching models of using information of job seekers' perceived probability of receiving an offer without conditioning on search effort ( 12,13 ), which is also used to motivate the functional form in theoretical search-and-matching models (3) .On the other hand, the rich heterogeneities and systematic patterns uncovered in our data suggest that more attention to behavioral responses amongst job seekers is warranted if these are key to any model's mechanism.
While respondents are over-optimistic about their job finding rate on average, we find that perceived returns to search satisfy several attractive properties.These beliefs correlate well with whether a worker received a job offer in the months following the first-wave survey: a 1 p.p. higher perceived job finding probability translates into a 0.2 p.p. higher probability of actually receiving an offer ( 2 = 0 .093 ).Higher target wages are associated with lower perceived job finding rates and receiving a job offer boosts the average expectation by around 8 p.p. in the second survey wave.Furthermore, we find that job finding probabilities are stationary despite the turbulences of the pandemic: workers perceive similar job finding rates when searching now compared to if they were instead searching in 6 months.
We also document systematic heterogeneities in perceived job finding probabilities and marginal returns to search.Respondents who are optimistic about the future state of the economy perceive the probability of being offered a job as significantly higher than those who think the economy will not be doing better going forward.However, optimistic respondents do not perceive the returns to delayed job search as significantly higher than returns to searching immediately.This suggests that the expectations about the state of the economy that we measure are capturing a broader optimism that influences workers' beliefs about their job finding rate.Moreover, women report higher returns to job search on average, and also perceive the marginal return to each extra hour spent searching for a job as higher.Lastly, we find evidence of significant associations between several personality traits and perceived job finding probabilities: individuals who are more willing to take risks, have higher confidence in their ability and feel that they are in control of their own life report significantly higher returns to search.
We find that the novel aspects introduced by the pandemic have had a relatively limited impact on workers' motives for search.The vast majority of on-the-job search is motivated by a desire to move into a more stable position or to increase wages, with 77% and 48% of workers, respectively, citing these two reasons as main determinants of their job search.A desire to work more from home was cited as a primary reason for searching by 14% of workers and only 5% cited that risk of Covid infection was a key motivation.Moreover, search behavior amongst employed and furloughed workers is uncorrelated with their ability to work from home.Thus, we find little evidence that pandemic-specific motivations are particularly salient in driving job search decisions.
This paper contributes to three main strands of the literature.First, we provide new evidence on job search behavior during the pandemic ( [4,14-16] ).This literature has shown that job search has responded to changes in the sectoral and occupation prevalence of job vacancies that has occurred since the start of the pandemic.Our results suggest that job search has not, however, been significantly altered by the new salience of working from home nor Covid-19 infection risk.More broadly, we contribute to the literature that has examined differences in job search strategies and behavior across demographic groups and personality traits. 2 Second, we provide new evidence on beliefs about the returns to search and the shape of the job-finding function using hypothetical vignettes.The use of hypothetical vignettes allows us to address the challenges that arise when using realized search behaviour and perceived returns given the endogeneity of behaviour and reservation wages to beliefs ( 11,27 ).Closest to our paper is [11] who analyze data from the Krueger-Mueller survey (12) to infer the perceived return to searching an additional hour beyond a job-seeker's chosen search intensity.While asking for the perceived return to an additional hour of search allows to study marginal returns to search, cross-subject comparisons need to consider that most respondents will be operating at different margins.We complement this work by eliciting job-finding probabilities at a set of multiple search hours, and analyzing whether perceptions between current and future returns differ and how perceived returns vary with target wages.Moreover, our survey allows us to study richer dimensions of heterogeneity in perceived returns including personality traits, which have been documented as important factors in relation to job search.
We elicit the elasticity of job finding rates with respect to effort, and find that workers perceive a comparably low return to additional hours searched.This finding lends support to the common practice of assuming constant job finding rates.Furthermore, in most cases, it has been found that job seekers are overoptimistic about their job finding rates, especially the long-term unemployed ( [13,27-29] ).Our detailed and interpersonally comparable measures of perceived job finding rates also confirm that individuals tend to perceive the average probability of being offered a new job as excessively high, and that out-of-work respondents are significantly more optimistic than those who are in paid work.These beliefs contrast with the "negative duration dependence ", i.e. the adverse effect of increasing unemployment spells documented by [30] using fake CV's in a field experiment, which is also a typical feature of search models (e.g. 31 ).
Finally, our work contributes to the literature that directly measures beliefs and examines their role in explaining actual behaviors.Existing work has, for example, made use of hypothetical scenarios to measure beliefs about the returns to different types of parental or educational investments ( [32-34] ) and maternal labor supply intentions (35) .To the best of our knowledge, ours is the first paper to collect interpersonally comparable measures of the perceived job finding function. 3Our findings are consistent with previous work showing that subjective probabilities vary meaningfully with observable characteristics and that individuals are typically aware of differential risks, even if the level of the probabilities they state is biased (37) .
The rest of this paper is organized as follows.Section 2 describes our data collection, analysis sample, and survey modules in detail.Section 3 characterizes search behavior, search effort, and the motivations for job search.Section 4 provides a detailed exploration of individual perceived returns to search effort including the extent to which these correlate with realized search behavior and job offers.Section 5 concludes with a discussion of the wider significance of our results.

Data
To study job search behavior and the role of perceived returns to search, we collect novel survey data from a large representative sample of adults in the UK.The data collection was carried out in two waves. 4he first wave of data was collected in April 2021, and the follow-up wave was carried out over July and early August 2021.To participate in the first wave of our study, respondents had to be residents in the UK, aged between 18 and 60 and have been in paid work for at least one month since January 2020.We collect 4,173 responses from the first wave, of which 3,000 from respondents who are currently in paid work and 1,173 from respondents who are out-of-work.We use quota-based sampling separately for in-work and out-of-work respondents to reach representativeness across age groups, educational attainment and broad regions in the UK conditional on employment status. 5 In the second wave, we have information on 2,556 respondents, which corresponds to a 61% re-contact rate. 6able A.1 compares the characteristics of respondents to our first and second survey wave, separately by employment status, to the distribution in the population of in-work and out-of-work adults in England.The distribution of in-work respondents to our first survey wave across educational attainment and regions matches the population distribution.Moreover, with our quota-based sampling we were able to closely match the share of respondents in the lowest age bracket (18-30 years old).For out-of-work respondents, we could not reach our intended sample size 3 Related work by Jäger et al. [36] analyzes biases in workers' beliefs about their outside options. 4Both surveys were conducted online by professional survey companies and designed to be accessible from all mobile devices and computers.We used the survey software Qualtrics to script the surveys.The median time to complete the survey was about 13 minutes for the first and 5 minutes for the second wave.Participants were paid modest incentives to take part in each of our survey wave, so payment for participation in the first survey was independent of whether they participated in the follow-up survey or not.We screen out participants to the first survey wave who fail an attention check and / or whose response time was too low.
5 Specifically, we match the joint distribution of the following characteristics, separately for in-work and out-of-work respondents: age above or below 30, having a university degree or not, and being resident in London and the South East, the rest of South of England and Midlands, or the North of England.Population-level estimates of the joint distribution of these characteristics in the population come from the third quarter of the 2020 Labour Force Survey (LFS). 6In the second wave, we managed to re-contact 1,961 and 595 respondents who were in work and out of paid work, respectively, at the time of our first survey wave. in the first wave and ended up oversampling young adults and individuals with a university degree.As a robustness check, we construct survey weights to match the joint distribution of our target variables in our sample to the distribution in the population and show in Appendix B that all results remain very similar. 7Despite not reaching the target sample size, the distribution of out-of-work respondents across regions in England closely matches population figures.In our follow-up survey, we managed to re-contact relatively older individuals and relatively more respondents with a university degree, while the distribution of wave-2 participants across regions in England is very close to the LFS figures.In terms of characteristics that we did not target with our sampling strategy, we note that our respondents are more likely to be female and married, compared to LFS participants.
In the following, we restrict our analysis sample to respondents who reported reservation wages below £10,000 per month in the first survey wave. 8Our final analysis sample thus includes 3,955 respondents for the first survey wave, and 2,437 for the second survey wave.Table A.2 presents summary statistics for our analysis sample, as reported at the time of the April data collection, separately by survey wave.Respondents to our April survey are on average 39 years old, 64% of our participants are women and 36% have at least one child.The average individual income from 2020 is around £24,000.Around half of our respondents were employed in April 2021, and 10% were either on partial or full furlough.Around 28% of our sample was out of paid work in April 2021.Looking at the summary statistics for respondents to our second survey wave, we can see that, on average, older and male respondents were more likely to participate in our second wave.On the contrary, we do not observe systematic differences in attrition based on employment status in April, income or presence of children.

Survey design
We here briefly describe the modules included in our surveys and how we elicit respondents' perceived returns to job search strategies.The full questionnaire can be found in Appendix C .
Employment status and job characteristics In our surveys, we ask respondents whether they were employed, self-employed, on partial or full furlough, or not in work in the week before the data collection.In-work and furloughed respondents are then asked detailed questions about their main job.Respondents who are out of work at the time of the data collection are asked similar questions about their last main job.
Expectations In the first survey wave, we elicit respondents' expectations about various aspects related to the evolution of the Covid-19 pandemic.In particular, we ask respondents whether they expect the economy to do better, the same, or worse in September 2021 than at the time of the data collection in April 2021.We use this question to construct a binary indicator for optimism, which takes the value one if the respondent thinks the economy will be doing better in September, and zero otherwise.
Job search behavior We obtain comprehensive information on respondents' intended and actual job search behavior and their job search effort.In the first survey wave, we ask respondents whether they are currently looking for a job, not currently looking for a job but planning to start within the next 12 months, or not planning to start looking for a job within the next year.Respondents who state they are currently looking for a job are asked when they started searching, how many hours per week they spend on job search on average, and the type of job they are looking for (which occupation, whether full-time or parttime jobs).Respondents who report planning to start searching for a new job within the next year are asked similar questions about their intended job search strategy.Respondents who report not planning to look for a job are asked to state the main reason for their choice.In the second survey wave, we ask respondents whether they had been looking for a job since April 2021 and, if so, how many hours they on average spent per week searching for a new job.
Perceived returns to job search To motivate the design of our survey instrument on perceived returns to search, we outline a simple directed model of job search following [3] . 9 In each time period  , workers choose their search effort,   , and reservation wage,   , to maximize their discounted expected utility.Search costs, (  ) , are increasing and convex and flow utility from consumption is given by the increasing, concave function  ( ⋅) .The reservation wage is the area of the wage distribution where workers target their search and can be thought of as akin to the desired wage level entered onto an online job finding platform.
At each period, the value functions of an unemployed worker,   (  ) , and employed workers,   (   ) , can be represented as: (1) where  are unemployment benefits,  gives the job destruction rate for employed workers, and   (   ,   ) gives the perceived job finding rate given search effort   at reservation wage   for a worker with employment status  .The characteristics of the function   (   ,   ) are a main object of interest in this paper.They determine a worker's optimal search intensity and their target wage.Rather than structurally estimate these functions on the basis of observed job finding rates and search patterns, we make use of hypothetical questions to elicit respondents' perceived returns to job search.Specifically, we present respondents with different hypothetical scenarios and ask what they think the likelihood is that they will find a new job within the next month.Motivated by the theoretical framework, the scenarios vary along the following three dimensions: (i) the reservation wage, (ii) the timing of the job search, and (iii) the search effort, expressed in terms of hours per week dedicated to job search.
Before we introduce the hypothetical scenarios, all respondents are asked to state what would be the lowest gross monthly salary that they would be willing to accept if they were offered a new job the following month (henceforth we refer to this self-reported minimum salary as the 'reservation wage').We then use their answers to construct two levels of wage: (1) a low wage, corresponding to their reported reservation salary, and (2) a high wage, which is 20% higher than the reported reservation salary.For each wage level, respondents are asked to think about the case where they start looking for a job immediately ('search now'), or in September 2021 ('search later').Finally, for each level of reservation wage and time of search, we ask respondents to think about the situation where they look for a new job for a month, and additionally vary the average number of hours per week spent searching for a new job ('search effort').For each scenario, we ask respondents for their perceived likelihood of finding a new employment within a month when they spend an average of 5, 10 or 15 hours per week, every week, searching for a job.Therefore, each respondent is presented with twelve scenarios, which we illustrate in Table 1 . 10espondents' answers to the twelve scenarios allow us to infer their beliefs about the level and shape of the job-finding function   (   ,   ) , which we assume to depend on search effort, reservation wage, and search timing.More specifically, we can learn something about how respondents' think their job finding probability depends on the (perceived) state of the labor market by comparing the self-reported likelihoods of finding a new job for the case where respondents immediately start searching to equivalent responses for the case where they postpone their search until September 2021.Additionally, by comparing answers to equivalent scenarios with different search intensities, we can draw inference on the degree of concavity or convexity of the job-finding function with respect to search effort.Similarly, we can compare answers to scenarios that only differ in the reservation wage to learn about the perceived elasticity of the job-finding function with respect to the desired salary.
Outcome of job search In our second survey wave, we elicit information on the outcome of respondents' job search from respondents who state they have been looking for a job since April 2021.In particular, we ask participants how many applications they have sent, interviews they have been contacted for, and offers they have received since April 2021.Additionally, we collect information on the type of job (full-time or part-time), occupation, and monthly salary of the three best offers that participants have received since April 2021, as well as whether they accepted any of these offers.
Personality traits and preferences In our first survey, we collect detailed information on personality traits and economic preferences.In particular, we use a standard 10-item module to measure respondents' locus of control [38] .We also administer experimentally validated survey modules to measure patience, willingness to take risks, confidence in one's own ability and the Big Five personality traits ( [39-42] ).

Search behavior
In this section, we describe the nature of the job search process for respondents to our April survey wave.We further discuss the reasons for on-the-job search, the intensity of search effort, and the job characteristics that individuals are targeting in their search.
Extensive margin Figure 1 gives the share of respondents who, in April 2021, were searching for a new job, not currently searching but planning to search within the next year, and neither searching nor planning to search by employment status.We see that the employed and selfemployed are similarly unlikely to be searching for a job with 11% and 13% currently searching, and 22% planning to search in the next year.The furloughed are slightly more likely to search, with 17% actively searching and 37% planning to search, which does not come as a surprise as the furlough scheme was scheduled to end by September 2021.Amongst those out of work, we see a much greater propensity to search for a new job, with 54% currently searching and 30% planning to do so within a year.
We can exploit the richness of our data to shed light on the reasons why workers who are currently in paid work may be looking for a new job.85% of in-work respondents are looking for a job to replace their current one, as opposed to a job in addition to their main employment.In Figure 2 we plot the share of in-work respondents that cited  different reasons for actively searching or planning to search for a job, separately for furloughed and not furloughed workers.The vast majority of searchers are motivated by a risk that their current position may end and that the pay in their current position is unsatisfactory.
Somewhat surprisingly, the risk of contracting Covid-19 at work or a desire to work more from home are not common reasons for job search for either furloughed or in-work respondents: approximately 5% of workers overall cite Covid risk as a driver of their job search, and less than 20% of workers report a desire to work more from home as one of the main reasons for their job search.In Table A.3 we explore demographic heterogeneities in whether workers report pandemic-related concerns as the main drivers of their job search.Overall, we do not uncover large systematic differences in whether job-searchers are motivated by Covid risk or teleworking possibilities (both with and without occupation and industry fixed effects).However, we find that women are 5 p.p. more likely to be citing working from home as a reason for job search.11This is consistent with [8] who find that US women in June 2021 were more likely to consider switching to a new job at the same pay if it allowed them to work two to three days from home per week.
We further check whether working from home is a salient driver of job search by analyzing the relationship between the ability to work from home in one's current job and the proportion searching.Figure 3 shows that the relationship between individuals' ability to do their tasks from home in their current or last job and search intentions is essentially flat.Thus, the precarity of one's current job is the main reason Fig. 3. Job search and ability to work from home Notes: The figure plots the relationship between the average share of tasks that workers can do from home in their main or last job (  -axis) and the share of workers who report being searching or planning to search for a job.The different panels show this relationship separately for in-work employees or self-employed workers (left), employees on partial or full furlough (center) and out-of-work respondents (right).The red line shows the line of best fit.Each bubble represents an occupation, and the size of the bubble is proportional to the number of observations in each occupation.The sample is restricted to respondents to the first survey wave and cells for which we have at least 5 observations.for searching for new employment, rather than the novel aspects of job design made salient by the pandemic.
Intensive margin Not everybody who is searching for a job searches with the same intensity.In Figure 4 , we show the distribution of the number of hours spent searching per week, separately by employment status.We see that, irrespective of employment status, the great majority of those searching non-zero hours in April 2021 searched between 1 and 5 hours in the week prior to data collection.Appendix Figure A.1 makes clear that, conditional on searching, those out of work search more.Looking at the average hours searched by employment status in the left panel of Figure 5 , we see that employed workers searched on average 5.5 hours per week, against a corresponding figure of 8.7 hours for those out of work.In Section 4 we consider whether these relatively low search intensities can be rationalized by low perceived returns to additional effort on the intensive margin.
Characteristics of job There is heterogeneity in the desired characteristics of jobs across job seekers.In particular, we elicit the minimum monthly salary that workers would be willing to accept when offered a new job, to shed light on the type of job workers are searching for.In the right panel of Figure 5 we see monthly reservation wages by employment status.The distributions are log-normal, and the distributions for the employed and self-employed are shifted to the right relative to the furloughed and those out of work.On average, out-of-work respondents would be willing to accept a monthly salary (before taxes) of £1358, against a reservation wage of about £2300 for in-work employees.
Lastly, individuals can target different occupations in their job search.The Covid-19 pandemic has highlighted the advantages of being able to perform one's tasks from the home office ( 43) , which may lead some workers to switch occupations in search of a job with more opportunities for telework.We again consider whether job seekers systematically target occupations to satisfy a desire to work more from home.The red circles and line in Figure 6 graphically represent the relationship be-tween the share of tasks that workers report being able to perform from home in their current or last job (x-axis) and the share of workers who report searching for a job exclusively in an occupation that is different from their current or last one, separately by employment status.Whilst individuals who are in paid work and who can do relatively few of their tasks from home report a significantly higher willingness to switch occupations compared to in-work or furloughed respondents who can do most of their tasks from home, the correlation between ability to work from home and desire to change occupation is less strong for those who are out of work. 12owever, the strong association between teleworking possibility and desire to switch occupation for in-work and furloughed respondents is largely driven by the overall less favorable job characteristics of jobs that cannot be done from home.As exemplified in Figure 2 , the main reasons for search on the job, in general, are a risk of the job ending and unsatisfactory pay.The gray circles and line in Figure 6 demonstrate that the positive association between the desire to switch occupation and the ability to work from home in one's existing job becomes insignificant once the income, permanency, and hours of the existing job are controlled for. 13Thus, the association between the ability to work from home and the desire to switch occupations is primarily driven by the fact that jobs which cannot be done from home tend to be unsatisfactory on other dimensions.This again suggests that job characteristics   The figure plots the relationship between the average share of tasks that workers can do from home in their main or last job (  -axis) and the share of workers who report being searching or planning to search for a job in a different occupation from their current or last one (red circles).In gray circles we represent the relationship between residualized willingness to switch occupations and ability to work from home.Residuals come from an OLS regression of whether a worker reports searching or planning to search for a job in a different occupation from their current or last one on log income, a dummy for whether the job is permanent, and the number of hours.Each bubble represents an occupation, and the size of the bubble is proportional to the number of observations in each occupation.The dashed lines represent the best linear fit.The different panels show this relationship separately for in-work employees or self-employed workers (left), employees on partial or full furlough (center) and out-of-work respondents (right).The sample is restricted to respondents to the first survey wave who report being actively searching for a job or planning to search within the next 12 months and cells for which we have at least 5 observations.made more salient by the pandemic are not the primary influence on individuals' job search strategies.14

Perceived returns to search
How do respondents perceive the returns to different job search strategies?We here use our elicited measures of perceived returns to job search to analyze how respondents believe their probability of receiving a job offer varies with search intensity, search timing, and targeted wage, and we identify systematic heterogeneities in perceived returns across background characteristics.
We start by examining our main measure of returns to job search, i.e., the perceived probability of receiving a job offer within one month, which we elicit using the hypothetical scenarios described in Section 2.1 .In Figure 7 , we use observations from the first survey wave in April and show perceived probabilities of receiving an offer, ( ⋅) , at the respondent's reservation wage (solid line) and when the desired wage is 20% higher (dashed line) for different levels of search intensity (x-axis).The left panel shows the perceived probabilities if searching now, whereas the right panel shows the corresponding probabilities if only beginning to search later (in September 2021). 15e note three features that emerge from the graphs.First, the returns to search effort, measured in hours per week spent searching for a job, look constant across the intensive margin of search. 16While respondents perceive large returns from searching 5 hours per week, the function appears linear with respect to search intensity, and the slope does not appear to be steep.For instance, the average perceived probability of receiving an offer is 32% at 5 hours searched per week, which would correspond to 6.4 p.p. per hour searched.However, the perceived probability at 15 hours of search is 46%, i.e. for each of the additional 10 hours spent searching, the perceived probability only increases by 1.4 p.p.This is true for both the high and low level of wages and irrespective of the timing of job search.
Second, perceived probabilities of receiving an offer are negatively correlated with the level of the desired wage: a target wage that is 20% above the reservation wage leads to an approximate drop of 5 p.p. in the average perceived probability of receiving a job offer, and the gap remains fairly constant across levels of search intensity.This is reassur-Fig.7. Average perceived probability of receiving an offer by hours searched.Notes: The figure shows average perceived probabilities of receiving a job offer at the respondent's reservation wage (solid lines) or when the reservation wage is 20% higher (dashed lines), by hours spent searching.The left panel shows these perceived probabilities for the scenarios where job search happens at the time of data collection (April 2021), whereas the right panel shows equivalent perceived probabilities for the scenarios where respondents only start looking for a job in September 2021.The thin black lines represent the 95% confidence intervals.The sample is restricted to respondents to the first survey wave.
ing as one would be unable to rationalize respondents' stated reservation wage if there was a weakly positive relationship between the probability of receiving a job offer and the desired wage.
Third, the perceived probability of finding a new job is uncorrelated with the timing of the job search.If anything, searching for a new job in September is thought to result in a lower average probability of being offered a job for all levels of search effort on average. 17However, perceptions about the state of the economy might influence how individuals perceive the returns to job search when searching at different times.In Figure A.5 we present average perceived probabilities of receiving an offer for different levels of search intensity and reservation wage, and with different search times, separately by whether or not the respondent believes the economy will be doing better in September 2021 compared to April (red lines), or not (blue lines).We define the former group as 'optimistic' respondents, and the latter as 'not optimistic' or 'pessimistic'.Overall, respondents who are optimistic about the economic outlook also perceive the probability of being offered a new job as significantly higher compared to pessimistic respondents, for all search intensities and levels of desired wage.

Predictors of perceived average returns
In Table 2 we analyze the determinants of the average perceived probability of receiving a job offer by means of ordinary least square regressions.In all columns, the dependent variable is the individual unweighted average perceived probability of receiving a job offer within one month, which we calculate by taking the mean of the perceived probabilities per hour searched across the twelve hypothetical scenarios (i.e., the non-optimized, hypothetical search choices): The sample is restricted to wave-one respondents.The average perceived probability of finding a job is 3.6% per hour searched with a considerable standard deviation of 2.4%. 18Column 1 shows that respondents perceive the likelihood of receiving an offer about 0.04 p.p. lower for every year of age, and respondents with a university degree perceive it 0.25 p.p. higher.Surprisingly, furloughed workers and individuals who are currently out of paid work perceive the probability of finding a new job to be 0.25 p.p. and 0.40 p.p. higher compared to the baseline group of workers who are in paid work at the time of data collection.These associations largely survive controlling for individual personality traits and region fixed effects (column 2), occupation and industry fixed effects (column 3) and the reservation wage (columns 4 and 5). 19ontrolling for the full set of individual characteristics, women still report 0.25 p.p. higher average job finding probabilities than men.The regression analysis further complements the finding from Figure 7 by showing that the perceived probability of receiving a job offer is decreasing in the reservation wage also across respondents.Turning to personality traits, respondents who are more patient, more willing to take risks and have higher confidence in their abilities also report significantly higher average perceived probabilities of being offered a new job.Amongst the Big Five traits, agreeableness, extraversion and openness to new experiences are all positively correlated with average perceived probabilities of being offered a job.In particular respondents with a 1 standard deviation higher agreeableness score perceive the job finding probability to be 0.13 p.p. higher.Lastly, in line with Caliendo et al.  [44] , feeling in control of one's own life is also found to be positively correlated with average perceived returns to job search.Notes: OLS regressions.Robust standard errors in parentheses.*  < 0 . 1 , * *  < 0 .05 , * * *  < 0 .01 .The sample includes respondents to first survey wave.The dependent variable is the perceived probability of receiving a job offer averaged across the twelve hypothetical scenarios, on a 0-100 scale.Region fixed effects refer to regions in England.

Predictors of perceived marginal returns
We now turn to examining individual perceptions about the return to an additional hour spent on job search.To obtain a measure of marginal returns, we calculate the difference in perceived probabilities of receiving an offer when searching 5 or 10 hours per week, and 10 or 15 hours per week, separately for the scenarios with different wage levels, and with immediate or postponed timing of search.We then average all these quantities and divide the average by 5 to obtain the individual specific average return to the marginal hour spent searching for a new job.On average, respondents believe that each additional hour per week spent on job search increases the probability of receiving a job offer within a month by 1.32 p.p. Figure 8 shows the relationship between marginal returns to one hour of search and the desired reservation wage: whilst there is evidence of a negative relationship between the reservation wage and perceived marginal returns to job search, the marginal returns are relatively flat across the distribution of reservation wages.
In Table 3 we regress the average perceived probability of receiving a job offer within a month on the number of hours spent searching in the relevant hypothetical scenario.We do this exercise separately for the scenarios with low and high levels of wage, and with different timings of job search.In all specifications, we control for individual fixed effects.We start with regressions that only include the number of hours spent searching in the hypothetical scenario and respondent fixed effects (columns 1, 3, 5 and 7).Perceived marginal returns to one additional hour searched are very similar across scenarios, and between 1.21 and 1.41 p.p. on average.
We then subsequently add interaction variables of hours spent searching with age, gender, whether the respondent has a university education and an indicator for the perceived state of the economy, to analyze how perceived marginal returns vary across background characteristics (see columns 2, 4, 6 and 8).In the baseline scenario, perceived returns to one additional hour of job search are 0.024 p.p. lower per year of age.The cross-sectional results for average returns are also confirmed for women, who perceive the marginal return to be 0.2 p.p. higher.For university educated individuals we find no significant difference in perceived marginal returns.Individuals who are optimistic about the future economic outlook also perceive the marginal return to search to be up to 0.33 p.p. higher.Given that this difference is independent of time of search, the individuals appear to be optimistic in general rather than just optimistic about the economy at a certain point in time.In Appendix Table A.4 we run the regressions separately for the employed, furloughed, and unemployed and find that the unemployed perceive the marginal returns as highest, followed by the furloughed, and finally the employed who have the lowest perceived marginal returns to an hour searched.In Appendix Table A.5 we further look into whether perceived marginal returns on average are linear by interacting hours searched with a dummy indicating the high intensity search scenario with 15 hours.The interaction coefficient is mildly negative and  insignificant in all specifications, suggesting that, if anything, perceived marginal returns on average are mildly concave.

Heterogeneity in perceived returns
We further look at the systematic heterogeneity in perceived returns across individuals by exploiting the fact that we have 12 scenarios per respondent.In Figure 9 we show the distribution of the different components of the parameterized search function recovered from running separate regressions for each individual explaining job finding probabilities. 20It is clear that there is substantial heterogeneity in the shape of the returns to search effort function across individuals; for example, some workers perceive searching later will raise their job finding probability, while others think theirs will be depressed.In Appendix Table A.6 we then relate estimated parameters of the search function to individual characteristics and find that risk takers perceive levels and marginal returns to be higher and that a higher locus of control predicts higher perceived marginal returns to job search. 21s a final step in documenting the shape of the job search function, we investigate whether the assumption of linearity in perceived returns at the intensive margin, i.e. constant marginal returns for search between 5-15 hours, is reasonable.For this endeavour we again run a regression for each individual separately.However, instead of controlling for hours searched, we control for medium (10h) and high (15h) search efforts using dummy variables, with a low (5h) search effort as the baseline category.We then compute the implied marginal return per hour for 5-10 hours searched and 10-15 hours searched.In the left panel of Figure 10 we see that distributions of marginal returns to one additional hour searched for medium vs high effort are very similar.Marginal returns for intermediate hours searched are slightly higher, which is confirmed in the scatter plot in the right panel.Therefore, it appears that the perceived job search function exhibits a mild concavity for most respondents.In Appendix Table A.7 we additionally interact medium and high intensity search efforts with individual characteristics and find that concavity is increasing in age and men also are more likely to perceive larger decreases in marginal returns than women.
Different occupations and industries in general face different realities, and particularly during the pandemic.We, therefore, in Figure 11 break down perceived job finding probabilities by occupation and industry.The occupations with the highest perceived marginal returns are 'Building and Grounds Cleaning and Maintenance' and 'Food Preparation and Serving', while those with lowest are 'Installation, Maintenance, and Repair' and 'Construction and Extraction'.It appears that perceived marginal returns are higher in some occupations such as 'Food Preparation and Serving' which had been hit hard by the pandemic and indeed were likely facing many vacancies while the economy was reopening.Similar patterns can be observed for industries on the right.In Appendix Figure A.7 , we further look at whether the perceived convexity or concavity varies systematically across occupations and industries by plotting the median ratio between the perceived marginal returns of hours 10-15 to 5-10.For all occupations and industries the median ratio lies below but relatively close to 1.

Do beliefs predict receiving a job offer?
A natural question that arises is to what extent beliefs about the shape of the job finding function predict respondents' job search strategies.Moreover, are respondents who perceive the probability of receiving an offer as higher, more successful in their job search?
In Figure 12 we plot the relationship between average perceived probabilities of receiving a job offer from the hypothetical scenarios (x-axis) and search behavior in April (y-axis).Respondents who think it is more likely on average that they will be offered a new job, holding constant search effort, are also more likely to be actively searching for a job at the time of the first data collection, or to report they are planning to start looking for a job within the next year.In Table 4 we confirm this unconditional association between perceived probability of finding a job and actual search behavior in a regression framework.The dependent variable is a binary indicator for whether the respondent reports either being actively searching for a job or planning to start searching within 12 months, at the time of the first data collection.The sample  is restricted to April respondents.As can be seen in column 1, controlling for employment status, a 10 p.p. increase in the average perceived probability of finding a job is associated with a 4 p.p. higher likelihood of being searching or planning to search for a new job.The coefficient decreases slightly but remains significant when additionally controlling for background characteristics (column 2), the respondent's reservation wage and region fixed effects (column 3), personality traits and optimism about the state of the economy (column 4), and occupation and industry fixed effects (column 5).When controlling for perceived returns to job search, we find a negative association between having an internal locus of control and being actively searching or planning to search for a job.This suggests that the positive association between locus of Fig. 12. Perceived probability of receiving an offer and search behavior.Notes: Binned scatter plot.The graph shows the relationship between respondents' average perceived probability of finding a job in April 2021 (  -axis) and whether respondents was actively searching for a new job in April.The red line represents the line of quadratic fit.The sample is restricted to respondents to the first survey wave.

Fig. 13.
Perceived probability of receiving an offer and outcome of job search.Notes: Binned scatter plots.The left panel shows the relationship between respondents' average perceived probability of finding a job elicited from the hypothetical scenarios in April 2021 (  -axis) and whether respondents actually received at least one offer between April 2021 and the time of the second data collection (  -axis).The right panel plots a similar relationship, but where on the  -axis we plot the number of offers received between April 2021 and the time of our second survey wave.In both panels, all variables have been residualized on month of interview fixed effects before plotting.The red line represents the linear fit.The sample is restricted to respondents to the second survey wave.control and job search that has been documented in previous literature (44) operates through differences in perceived returns to job search.
Next we turn to analyzing whether differences in perceptions about the impact of job search are also associated with differences in job search success.In Figure 13 we exploit information from our second survey wave on the outcome of the search process to show that average perceived probabilities of being offered a new job elicited from the hypothetical scenarios are positively correlated with the probability of receiving at least one job offer since April (left panel) and the actual number of job offers received (right panel) between April 2021 and the time of the second data collection.Table 5 further confirms this positive association between perceived job finding probability and search success in a regression framework.Our specifications that control for a broad range of individual and job characteristics, as well as search intensity, show that respondents who report a 10 p.p. higher average probability of being offered a job are also 1.4 p.p. more likely to have received a job offer (column 2), and have received 0.04 more offers on average (column 4).

Updating
Finally, we exploit information from our second survey wave to analyze how respondents update their perceived returns to hours spent searching for a job, and whether there are systematic differences in updating behavior depending on the outcome of one's job search. 22In Table A.8 we regress the average perceived job finding probability as elicited at the time of our second survey wave on the same average perceived probability calculated from responses to the first survey wave, as well as on two different measures of job search success.The analysis shows that respondents with higher perceived impacts of job search in April also report higher average job finding probabilities in July.Moreover, participants update the perceived impacts of job search upwards  2) is a binary indicator for whether the respondent received a job offer between April 2021 and the second survey wave.The dependent variable in Columns ( 3) and ( 4) is the number of offers received since April 2021.Region fixed effects refer to regions in England.
when successful with their job search. 23Indeed, column 1 shows that respondents who have received at least one job offer since April 2021 report a 8.1 p.p. higher average perceived job finding probability in wave 2. Similarly, average perceived returns to job search are higher, the more offers participants have received between April and July 2021 (see column 2).This includes Mueller et al. [13] who show that on average job seekers do not appear to update their beliefs about the probability of finding a job and Potter [29] who shows how job seekers adjust their job search 23 This is consistent with [29] who shows that search effort jumps discretely after a job offer is received in the context of the Great Recession.
behavior in response to the receipt of an offer, which is in line with the results in Appendix Table A.8 .

Conclusion
In this paper, we shed light on the perceived returns to job search effort.We show that while individuals are, on average, over-optimistic about their likelihood of receiving a job offer in the next month, subjective probabilities vary meaningfully with observable characteristics and correlate with realized search behavior and job finding rates.Individuals perceive a trade-off between job finding rates and the target wage and perceive a relatively low return to additional search effort beyond 5 hours per week.This contrasts with policy approaches where eligibility to benefits often hinges on treating job search like a full-time job.
We have also presented detailed evidence on the characteristics of job search in the UK economy as policy moved to emphasize adaptation to Covid-19 over virus elimination.We show that new aspects of job design made salient by the pandemic are not primary motives for onthe-job search.A desire to work more from home was cited as a primary reason for search by 14% of workers, while only 5% cited the risk of Covid infection.This is compared to 77% of workers who were searching due to a concern about their position ending.The probability of search amongst employed and furloughed workers is uncorrelated with their ability to work from home in their current job, and a preference for switching into a new occupation is uncorrelated with the ability to work from home once pay, hours, and the permanency of the existing contract are accounted for.
Our findings on the shape of the perceived job finding function have implications for empirical and theoretical work on job search.Most survey modules on perceived job finding probabilities do not condition on search effort.For example, the Survey of Consumer Expectations asks: "What do you think is the percent chance that within the coming 3 months, you will find a job that you will accept, considering the pay and type of work?" We find that respondents on average perceive a significantly positive but relatively low marginal return to search.This finding can, absent more detailed information, justify this common practice and can also help rationalize the relatively low search intensities reported in surveys.Furthermore, theoretical treatments of job search can calibrate their choice of functional form to be consistent with the perceived elasticities of job finding probabilities with respect to search effort documented in this paper.

Fig. A.2. to switch occupation and ability to work from home by gender. Notes:
The figure plots the relationship between the average share of tasks that workers can do from home in their main or last job (  -axis) and the share of workers who report being searching or planning to search for a job in a different occupation from their current or last one, separately by gender.The top and bottom rows show graphs for male and female respondents and female respondents, respectively.The different panels show this relationship separately for in-work employees or self-employed workers (left), employees on partial or full furlough (center) and out-of-work respondents (right).The red line shows the line of best fit.Each bubble represents an occupation, and the size of the bubble is proportional to the number of observations in each occupation.The sample is restricted to respondents to the first survey wave who report being actively searching for a job or planning to search within the next 12 months and cells for which we have at least 5 observations.Notes : Columns 1, 2 and 4, 5 shows the demographic characteristics of our sample, separately by survey wave and by employment status.Columns 2 and 6 show the corresponding weighted figures from the third quarter of the 2020 Labour Force Survey. ( ) These variables were not targeted in our quota-based sampling strategy.Notes : The table shows the background characteristics of our analysis sample, separately by survey wave.The sample is restricted to respondents to the first survey wave.The graph shows average perceived probabilities of receiving a job offer at the respondent's reservation wage (solid lines) or when the reservation wage is 20% higher (dashed lines), by hours spent searching and by whether or not the respondent thought the economy would be doing better in September 2021 than in April 2021 (red and blue lines indicate an optimistic or a neutral / pessimistic outlook, respectively).The left panel shows these perceived probabilities for the scenarios where job search happens at the time of data collection (April 2021), whereas the right panel shows equivalent perceived probabilities for the scenarios where respondents only start looking for a job in September 2021.The thin vertical lines represent 95% confidence intervals.Notes: OLS regressions.Standard errors in parentheses.*  < 0 . 1 , * *  < 0 .05 , * * *  < 0 .01 .The sample includes respondents to first survey wave.The dependent variable is the perceived probability of receiving a job offer (0-100).The different columns refer to the different hypothetical scenarios that vary in whether respondents are searching at the time of data collection (columns 1 and 2) or in September 2021 (columns 3 and 4), with the normal and high level of reservation wage respectively.In all columns, regressions are performed using the responses to the three relevant hypothetical job search scenarios.'Hours of search' takes values 5, 10 and 15, based on the relevant hypothetical scenario.All regressions include individual fixed effects.The different panels refer to regressions run separately on in-work respondents (top), respondents who are in partial or full furlough (middle), and out-of-work respondents (bottom).

Table A.5
Perceived marginal returns to search with individual fixed effects and interactions.

Fig. 1 .
Fig. 1.Search behavior by employment status.Notes: The figure shows the share of respondents to the first wave who are actively searching for a job, planning to start searching within the next 12 months or not planning to search, by employment status in the week before the data collection.The sample is restricted to respondents to the first survey wave.

Fig. 2 .
Fig. 2. Reasons for job search.Notes: The figure shows the share of in-work respondents who are actively searching or planning to search for a new job, who mention each of the possible reasons for searching.Multiple responses are possible.The left panel shows answers for inwork employees and self-employed workers, whereas the right panel shows answers for respondents who are either in partial or full furlough.The sample is restricted to respondents from the first survey wave who report being in paid work and being actively searching for a job or planning to search within the following 12 months.The lines represent the 95% confidence intervals.

Fig. 4 .
Fig. 4. Hours searched by employment status.Notes: The figure shows the share of respondents to the first survey wave who searched for a job for a certain number of hours by employment status in the week before the data collection.

Fig. 5 .
Fig. 5. Average hours searched and monthly reservation wage (in £) by employment status.Notes: In the left panel we present average hours searched in the last week conditional on searching by employment status.The thin bars represent the 95% confidence interval.In the right panel we display the kernel density of monthly reservation wages by employment status.The sample is restricted to observations from the first survey wave.

Fig. 6 .
Fig.6.Intentions to switch occupation and ability to work from home.Notes: The figure plots the relationship between the average share of tasks that workers can do from home in their main or last job (  -axis) and the share of workers who report being searching or planning to search for a job in a different occupation from their current or last one (red circles).In gray circles we represent the relationship between residualized willingness to switch occupations and ability to work from home.Residuals come from an OLS regression of whether a worker reports searching or planning to search for a job in a different occupation from their current or last one on log income, a dummy for whether the job is permanent, and the number of hours.Each bubble represents an occupation, and the size of the bubble is proportional to the number of observations in each occupation.The dashed lines represent the best linear fit.The different panels show this relationship separately for in-work employees or self-employed workers (left), employees on partial or full furlough (center) and out-of-work respondents (right).The sample is restricted to respondents to the first survey wave who report being actively searching for a job or planning to search within the next 12 months and cells for which we have at least 5 observations.
p  ,ℎ, 12 × 10 (3) 17 Figure A.4 shows that perceived probabilities of receiving an offer behave very similarly for in-work, furloughed and out-of-work respondents.

Fig. 8 .
Fig. 8. Perceived marginal return to hour searched by reservation wage.Notes: The figure shows the average increase in the perceived probability of finding a job for one extra hour spent searching, for different levels of the respondents' reservation wage (  -axis).The sample is restricted to respondents from the first survey wave.

Fig. 9 .
Fig. 9. Distributions of estimated coefficients of perceived job finding probability for each respondent.Notes: The Figure shows the distribution of estimated coefficients from the OLS regressions explaining job finding probability   ∈ [0 , 100] as in   =   +   ℎ  +     +     +  based on hours searched ℎ , timing  , and low or high wage  for each individual  across scenarios .The top left panel shows the histogram of individual coefficients α , the top right of β , the bottom left of γ , and the bottom right of ζ .The red dashed line indicates the median.The sample is restricted to respondents from the first survey wave and the plot excludes the top and bottom five percentiles.

Fig. 10 .
Fig. 10.Distributions of estimated coefficients for marginal returns to medium and high search efforts for each respondent.Notes: The Figure shows the distribution of estimated coefficients from the OLS regressions explaining job finding probability   ∈ [0 , 100] as in   =   +    ℎ  +    ℎ  +     +     +  based on dummy variables for medium (10 h) and high (15 h) hours searched ℎ , timing  , and low or high wage  for each individual  across scenarios .The top left panel shows the histogram of marginal returns to one hour search of medium vs. high effort.The right panel shows a scatter plot of to one hour search of medium vs. high effort.The black line represents the 45 degree line.The sample is restricted to respondents from the first survey wave and the plot excludes the top and bottom five percentiles.

Fig. 11 .
Fig. 11.Mean perceived marginal return by occupation and industry.Notes: The Figure shows the distribution of estimated coefficients from the OLS regressions explaining job finding probability   ∈ [0 , 100] as in   =   +   ℎ  +     +     +  based on hours searched ℎ , timing  , and low or high wage  for each individual  across scenarios .The figure displays the mean   across occupations (left) and industries (right).

Fig. A. 3 .
Fig. A.3.Distribution of perceived probability of receiving an offer by hours searched.Notes: The figure shows the distribution of average perceived job finding probabilities at the desired reservation wage from the hypothetical scenarios where respondents start their job search at the time of data collection, for different levels of search intensity.The sample is restricted to respondents to the first survey wave.

Fig. A. 4 .
Fig. A.4.Average perceived probability of receiving an offer by hours searched & employment status.Notes: The sample is restricted to respondents to the first survey wave.The graphs show average perceived probabilities of receiving a job offer by hours spent searching.Panel (a) shows average perceived probabilities for in-work respondents (either employees or self-employed).Panels (b) and (c) refer to furloughed and out-of-work respondents, respectively.The thin vertical lines represent 95% confidence intervals.

Fig. A. 6 .
Fig. A.6.Average perceived probability of receiving an offer by hours searched & employment status.Notes: The sample is restricted to respondents to the first survey wave.The graph shows average perceived probabilities of receiving a job offer, by hours spent searching and by employment status in week before the data collection.The thin vertical lines represent 95% confidence intervals.

Fig. A. 7 .
Fig. A.7. Median ratio between perceived return hours 10-15 and hours 5-10 across occupations and industries.Notes: The figure displays the median ratio of high intensity to medium intensity returns

Table 1
Overview of hypothetical scenarios.

Table 2
Predictors of average perceived probability of receiving an offer per hour searched.

Table 3
Predictors of perceived probability of receiving an offer with individual fixed effects.*  < 0 .05 , * * *  < 0 .01 .The sample includes respondents to first survey wave.The dependent variable is the perceived probability of receiving a job offer (0-100).Columns 1, 2 and 3, 4 refer to the scenarios where the respondents are searching at the time of data collection, with the normal and high level of reservation wage respectively.Columns 5, 6 and 7, 8 report equivalent regressions for the scenarios where respondents are only searching for a job in September 2021.In all columns, regressions are performed using the responses to the three relevant hypothetical job search scenarios.'Hours of search' takes values 5, 10 and 15, based on the relevant hypothetical scenario.All regressions include individual fixed effects.

Table 4
Predictors of search behavior.
Notes: OLS regressions.Standard errors in parentheses.*  < 0 . 1 , * *  < 0 .05 , * * *  < 0 .01 .The sample includes respondents to first survey wave.The dependent variable is a binary indicator for whether the respondent is either actively searching for a job in April 2021 or plans to start searching for a job within the following year.Region fixed effects refer to regions in England.

Table 5
Predictors of outcome of job search.

Table A .1
Sample representativeness.

Table A .2
Descriptive characteristics of the sample.

Table A .3
Risk of Covid and WFH as reasons for job search.*  < 0 .05 , * * *  < 0 .01 .The sample includes respondents to first survey wave who are in work and who are either searching or a job or planning to search within the next year.The dependent variables take value one if the respondent indicates the risk of contracting Covid or wanting to work from home more as the main reason to look for a job or being planning to search, and zero otherwise.Region fixed effects refer to regions in England.Perceived marginal returns to search with individual fixed effects by employment status .
Notes: OLS regressions.Robust standard errors in parentheses.*  < 0 . 1 , * < 0 . 1 , * *  < 0 .05,***  < 0 .01.The sample includes respondents to first survey wave.The dependent variable is the perceived probability of receiving a job offer (0-100).Columns 1, 2 and 3, 4 refer to the scenarios where the respondents are searching at the time of data collection, with the normal and high level of reservation wage respectively.Columns 5, 6 and 7, 8 report equivalent regressions for the scenarios where respondents are only searching for a job in September 2021.In all columns, regressions are performed using the responses to the three relevant hypothetical job search scenarios.'Hours of search' takes values 5, 10 and 15, based on the relevant hypothetical scenario.'Hours×15h'indicates an interaction between the 'Hours of search' variable and an indicator for the hypothetical scenario with highest search intensity (15 h).All regressions include individual fixed effects.OLS regressions.Standard errors in parentheses.*<0 . 1 , * *  < 0 .05,***  < 0 .01.The sample includes respondents to first survey wave.The dependent variables are coefficients from individual-level regressions.In these individual level regressions, the dependent variable was the perceived job finding probability from the hypothetical scenarios, and the explanatory variables where search effort and binary indicators for searching later and the higher level of target wage.Columns 1 and 2 have the constant from individual-level regressions as dependent variable, whereas successive pairs of columns use as dependent variable the coefficients on search effort, the binary indicator for searching in September 2021 and the binary indicator for the high level of wage.Predictors of perceived probability of receiving an offer with individual fixed effects and interactions between levels of search and individual characteristics.OLS regressions.Standard errors in parentheses.*<0 . 1 , * *  < 0 .05,***  < 0 .01.The sample includes respondents to first survey wave.The dependent variable is the perceived probability of receiving a job offer (0-100).Columns 1, 2 and 3, 4 refer to the scenarios where the respondents are searching at the time of data collection, with the normal and high level of reservation wage respectively.Columns 5, 6 and 7, 8 report equivalent regressions for the scenarios where respondents are only searching for a job in September 2021.In all columns, regressions are performed using the responses to the three relevant hypothetical job search scenarios.'10h of search' is a dummy indicating 10 h of search and and '15 h of search' is a dummy indicating 15 h of search based on the relevant hypothetical scenario.The baseline search intensity is 5 h.All regressions include individual fixed effects.<0.05,** *  < 0 .01.The sample includes respondents to second survey wave.The dependent variable is the average perceived probability of receiving a job offer, averaged over the six hypothetical scenarios (0-100).'Averageperceivedprobability in April' is calculated in a similar way from responses to the first survey wave.'ReceivedoffersinceApril'takes value one if the respondent reports having received at least one job offer since April 2021.'TotaljobofferssinceApril 2021' is the total number of job offers received between April 2021 and the time of the second data collection.Month fixed effects refer to month of interview for the second survey.OLS regressions.Robust standard errors in parentheses.Observations are weighted to match the joint population distribution of age, educational attainment and broad region of residence in England, separately for in-work and out-of-work respondents.*<0 . 1 , * *  < 0 .05,***  < 0 .01.The sample includes respondents to first survey wave.The dependent variable is the perceived probability of receiving a job offer averaged across the twelve hypothetical scenarios, on a 0-100 scale.Region fixed effects refer to regions in England.Predictors of perceived probability of receiving an offer with individual fixed effects -Weighted results.<0.01.Weights are constructed to match the joint population distribution of age, educational attainment and broad region of residence in England, separately for in-work and outof-work respondents.The sample includes respondents to first survey wave.The dependent variable is the perceived probability of receiving a job offer (0-100).Columns 1, 2 and 3, 4 refer to the scenarios where the respondents are searching at the time of data collection, with the normal and high level of reservation wage respectively.Columns 5, 6 and 7, 8 report equivalent regressions for the scenarios where respondents are only searching for a job in September 2021.In all columns, regressions are performed using the responses to the three relevant hypothetical job search scenarios.'Hours of search' takes values 5, 10 and 15, based on the relevant hypothetical scenario.All regressions include individual fixed effects.Predictors of coefficients from individual regressions explaining perceived probability of receiving an offer -Weighted results.OLS regressions.Standard errors in parentheses.*<0 . 1 , * *  < 0 .05,***  < 0 .01.The sample includes respondents to first survey wave.The dependent variables are coefficients from individual-level regressions.In these individual level regressions, the dependent variable was the perceived job finding probability from the hypothetical scenarios, and the explanatory variables where search effort and binary indicators for searching later and the higher level of target wage.Columns 1 and 2 have the constant from individual-level regressions as dependent variable, whereas successive pairs of columns use as dependent variable the coefficients on search effort, the binary indicator for searching in September 2021 and the binary indicator for the high level of wage.Weights are constructed to match the joint population distribution of age, educational attainment and broad region in England, separately for in-work and out-of-work respondents.OLS regressions.Standard errors in parentheses.Weights are constructed to match the joint population distribution of age, educational attainment and broad regions in England, separately for inwork and out-of-work respondents.*<0 . 1 , * *  < 0 .05,***  < 0 .01.The sample includes respondents to first survey wave.The dependent variable is a binary indicator for whether the respondent is either actively searching for a job in April 2021 or plans to start searching for a job within the following year.Region fixed effects refer to regions in England.Predictors of outcome of job search -Weighted results.OLS regressions.Standard errors in parentheses.*<0 . 1 , * *  < 0 .05,***  < 0 .01.The sample includes respondents to first survey wave.The dependent variable in Columns (1) and (2) is a binary indicator for whether the respondent received a job offer between April 2021 and the second survey wave.The dependent variable in Columns (3) and (4) is the number of offers received since April 2021.Region fixed effects refer to regions in England.Explaining average perceived probability of receiving an offer with past beliefs and offers -Weighted results.