Journal Of Applied Psychology - Harvard University

LEE, GINO, AND STAATSThis document is copyrighted by the American Psychological Association or one of its allied publishers.This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.2that they can be engaged in without explicit awareness, thoughthey are not directed toward the given task (Smallwood &Schooler, 2006). Thus, when the mind wanders, attention shiftsaway from the given task and may lead to failures in task performance (Manly, Robertson, Galloway, & Hawkins, 1999; Robertson, Manly, Andrade, Baddeley, & Yiend, 1997). Prior work notesthat general cognitive interference can have costly effects onworker productivity (for a review, see Jett & George, 2003).Workers who experience cognitive interference are distracted,showing an inability to focus on a task (Fisher, 1998) and a greaterlikelihood of committing errors while completing the task (Flynnet al., 1999).Thinking about salient and attractive outdoor options is a formof task-unrelated thinking that serves as a cognitive distraction thatshifts workers’ attention away from the task at hand. Accordingly,we expect it will be harder for workers to maintain their taskrelated thoughts on good weather days than on bad weather days.As a result, we also predict that workers will be less productive ongood weather days than on bad weather days. More specifically,we argue that on a bad weather day, individuals will have a higherability to focus on a given work task not because of the negativemood induced by the weather but because fewer distractingthoughts related to outdoor options will be readily available intheir minds. Consequently, they will be able to better concentrateon their tasks and work more productively on bad weather days. Inour research, we consider tasks where productivity requires highlevels of attention and focus, which allow workers to completetheir work faster. Thus, we expect fewer cognitive distractions tobe associated with higher levels of work productivity. Takentogether, these arguments lead to the following hypotheses:Hypothesis 1. Good weather conditions, such as lack of rain,will decrease worker productivity on tasks that require sustained attention and focus, compared to bad weatherconditions.Hypothesis 2. Good weather conditions will increase the salience and attractiveness of outdoor options, compared to badweather conditions.Hypothesis 3. The relationship between good weather conditions and worker productivity will be mediated by greatercognitive distractions (i.e., salience of one’s outdoor options).To test our predictions, we used empirical data on workerproductivity, measured by individual performance on tasks conducted in a Japanese bank (Study 1), an online marketplace (i.e.,Amazon Mechanical Turk, Studies 2 and 3), and the laboratory(Study 4). We focused on precipitation as the key measure of badweather given the previous finding that precipitation is the mostcritical barrier to outdoor physical activities (Chan, Ryan, &Tudor-Locke, 2006; Togo, Watanabe, Shephard, & Aoyagi, 2005).Study 1: Field Evidence From a Japanese BankMethodIn Study 1, we examined the proposed link between weatherconditions and productivity by matching data on employee productivity from a mid-size bank in Japan with daily weather data.1In particular, we assessed worker productivity using archival datafrom a Japanese bank’s home-loan mortgage-processing line. Forthe sake of brevity, we discuss the overall structure of the operations here; more detailed information can be found in Staats andGino (2012). Our data includes information on the line from therollout date, June 1, 2007 until December 30, 2009, a 2.5-year timeperiod. We examined all transactions completed by the permanentworkforce, 111 workers who completed 598,393 transactions.Workers at the bank conducted the 17 data-entry tasks required tomove from a paper loan application to a loan decision. Includedwere tasks such as entering a customer’s personal data (e.g., name,address, phone number) and entering information from a real estateappraisal. Workers completed one task at a time (i.e., one of 17steps for one loan); when a task was completed, the systemassigned the worker a new task. The building in which the worktook place had windows through which workers could observe theweather. Workers were paid a flat fee for their work; there was nopiece-rate incentive to encourage faster completion of work.In addition to the information on worker productivity, we alsoassembled data on weather conditions in Tokyo, the city where theindividuals worked. The National Climactic Data Center of theU.S. Department of Commerce collects meteorological data fromstations around the world. Information for a location, such asTokyo, was calculated as a daily average and includes summariesfor temperature, precipitation amount, and visibility.MeasuresCompletion time. To calculate completion time, we took thenatural log of the number of minutes a worker spent to completethe task ( 0.39, 1.15). As we detail below, we conductedour analyses using a log-linear learning curve model.Weather conditions. Since our main variable of interest isprecipitation, we included a variable equal to the amount of precipitation each day in inches, down to the hundredth of an inch( 0.18, 0.53). To control for effects from other weatherrelated factors, we also included temperature ( 62.1, 14.6)and visibility ( 10.3, 5.1). With respect to the former, itmay be that productivity is higher with either low or high temperatures. Therefore, we entered both a linear and quadratic term fortemperature (in degrees Fahrenheit). Finally, because worse visibility could be related to lower productivity, we included theaverage daily visibility in miles (to the tenth of a mile).Control variables. We controlled for variables that have beenshown to affect worker productivity. These included: same-day,cumulative volume (count of the prior number of transactionshandled by a worker on that day); all prior days’ cumulativevolume (count of transactions from prior days); load (percentageof individuals completing work during the hour that the focal taskoccurred; see Kc & Terwiesch, 2009); overwork (a comparison ofcurrent load to the average, see Kc &Terwiesch, 2009); defect;day-of-week, month, year, stage (an indicator for each of the 17different steps); and individual indicators.1The data reported in Study 1 have been collected as part of a larger datacollection. Findings from the data have been reported in separate articles:Staats and Gino (2012) and Derler, Moore, and Staats (2013).

This document is copyrighted by the American Psychological Association or one of its allied publishers.This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.BAD WEATHER INCREASES PRODUCTIVITY3Results and DiscussionMethodWe used a log-linear learning curve model because individuals’performance improves over time with experience. Using this approach, we conducted our analyses at the transaction level. Therefore, in our models, we controlled for the effects of the worker,task, and time, and then examined the effect of weather on workerproductivity. For our primary model, we used a fixed effects linearregression model with standard errors clustered by individual.Column 1 in Table 1 shows our main model, which we used totest Hypothesis 1. Examining rain, we found that the coefficient isnegative and significant (coefficient 0.01363). In terms of theeffect size, we found that a one-inch increase in rain is related toa 1.3% decrease in worker completion time for each transaction.Given that there are approximately 100 workers in the operation,a 1.3% productivity loss is approximately equivalent to losing oneworker for the organization on a given day. Based on the averageyearly salary of the associate-level employees at this bank and theaverage frequency of precipitation, this loss could cost approximately 18,750 for this particular operation a year. When accumulated over time for the entire bank of nearly 5,000 employees,a 1.3% productivity loss could be interpreted as a significant lossin revenue for the bank: at least 937,500 a year. Further, in a citythe size of Tokyo (approximately 9 million people) our identifiedeffect could translate into hundreds of millions of dollars in annuallost productivity.Next, it is important to properly account for the standard errorsin our model as we have many observations nested within a smallnumber of individual workers. Therefore, in Column 2, we clustered the standard errors by day, not by worker. In Column 3, weused Prais-Winsten regression with panel-corrected standard errorsadjusted for heteroskedasticity and panel-wide, first-order autocorrelation. Then, in Column 4, we used the fixed effects regressionmodel from Columns 1–3 but used block-bootstrapped standard errors.In each model, the coefficient on rain is negative and statisticallysignificant. Finally, in Columns 5 and 6 we added additionalcontrols with first individual fixed effects interacted with monthlyfixed effects and then individual fixed effects interacted with stagefixed effects. In conclusion, using a within-subject design, thisstudy showed that greater rain is related to better worker productivity.Participants and procedure. We recruited U.S. residents toparticipate in an online survey in early March, when weatherconditions vary significantly depending on where workers arelocated. Three hundred twenty-nine online workers (Mage 36.52years, SD 12.79; 48% male) participated in a 30-min study andreceived a flat fee of 1. We first gave all workers a threeparagraph essay that included 26 spelling errors; we asked them tofind as many errors as they could and correct the errors theyfound.2Once all the workers had completed the task, they completed aquestionnaire that included measures of state emotions to controlfor potential effects of affect. Finally, we asked workers to complete a demographics questionnaire that also included questionsabout the day’s weather and their zip code.Measures.Productivity. We computed the time (in seconds) workersspent on the task of correcting spelling errors (i.e., speed). Giventhat each worker spent a different amount of time on the task, wecalculated speed by dividing the number of typos detected by thetotal time taken in seconds. We then log-transformed the variableto reduce skewness. In addition, we computed how many spellingerrors were correctly identified and fixed as a measure of accuracy.State emotions. We used the 20-item form of the Positive andNegative Affect Scale (PANAS; Watson, Clark, & Tellegen,1988). Participants indicated how much they felt each emotion“right now” using a 7-point scale. We calculated two summaryvariables for each participant: positive ( .90) and negativeaffect ( .91).Weather questionnaire. Workers were asked to report theirzip code, which enabled us to find the daily weather data of thespecific area on a specific day (http://www.wunderground.com).To ensure that workers’ perceived weather matched actual weatherconditions, we also asked them to think about the weather conditions of the day, relative to their city’s average weather conditions,using a 5-point scale (1 one of the best to 5 one of the worst).Study 2: Online Study of Weather and ProductivityAlthough Study 1 offers valuable information on employees’actual work productivity, only the time taken to complete a taskwas used as an outcome variable, as error rates were low (less than3%) and showed little variation across employees. In Study 2, wesought a conceptual replication of the effect of weather on completion time while also using a task that would permit us tomeasure error rates. We could thus investigate productivity notonly in terms of quantity (speed at which workers completed theirgiven task) but also in terms of quality (accuracy of detectingerrors and correcting them). To account for the potential influenceof weather-driven moods, in addition to new productivity measures, we collected data on whether workers felt positive or negative affect while completing the task.Results and DiscussionWe first tested whether actual weather matched workers’ perceptions of the day’s weather. Indeed, subjective perceptions ofbad weather were associated with lower temperature (r .24,p .001), higher humidity (r .21, p 0.001), more precipitation (r .23, p 0.001), more wind (r .31, p 0.001), andlower visibility (r .26, p 0.001).Table 2 reports summary statistics. Table 3 summarizes a seriesof regression analyses. Consistent with Hypothesis 1, more rainwas associated with higher productivity, measured in terms of bothspeed and accuracy (Model 1). This relationship holds even aftercontrolling for key demographic variables and state emotions(Model 2). These findings suggest that bad weather is associatedwith both indicators of productivity, increased speed, and accuracy.2More detailed instructions and materials are available online as supplemental materials (Appendix A).

Bⴱ0.006068 0.013630.0043400.006964ⴱ3.710e 05 6.425e 05ⴱ7.311e 04 9.799e 04SEBⴱⴱⴱSEⴱBⴱ0.005686 0.011670.0043640.0045193.819e 05 4.588e 057.040e 04 8.176e 04SEBⴱ0.0060550.0044383.756e 057.102e 04SE6Individual Stage fixedeffects0.004827 0.013360.0034730.0068632.946e 05 6.449e 055.755e 04 7.808e 04SE5Individual Month —598,3930.3374——598,3930.3563—6.198e 11 1.524e 09ⴱ0.01030 06ⴱⴱⴱ0.08566 0.3350—598,3930.08806Yes7.360e 10 1.380e 090.05965 ⴱⴱⴱ0.23940.1733Yes598,3930.04908—5.905e 100.047880.041080.035000.21541.205e 09 1.323e 09ⴱ0.05141 ⴱⴱⴱ0.21601.0083ⴱⴱⴱ1.461e 10 1.508e 09ⴱⴱⴱ0.02195 ⴱⴱⴱ0

of North Carolina at Chapel Hill s Center for International Business Education and Research, and the University Research Council at the University of North Carolina at Chapel Hill. We thank Max Bazerman and Karim Kassam for their insightful comments on earlier drafts of this article.