The Economics of Telecommuting: Theory and Evidence

Transcription

1 The conomics of Telecommuting: Theory and vidence Guodong Gao The Wharton School University of Pennsylvania Lorin M. Hitt The Wharton School University of Pennsylvania Astract While there has een considerale research on the effect of telecommuting on worker s productivity and quality of work life, there is consideraly less work on the managerial prolems associated with selecting, monitoring and compensating workers involved in telecommuting. We propose a model ased on contract theory to analyze the managerial decisions on telecommuting, focusing on: How managers should decide which workers will have the opportunity to telecommute, and How managers should monitor and provide incentives to workers who participate in telecommuting programs. Based on the model, we find that managers willingness to allow telecommuting is related to the amount of information they have aout their employees and that employees who telecommute should have incentives ased oth on sujective evaluations and ojective measures. Using data from the 998 Workplace mployment Relationship Survey WRS98, we test these predictions and find that they are supported y the data. Key words: telecommuting, distriuted work arrangements, information technologies, multitasking agent, sujective measurement, contract theory I. Introduction The concept of telecommuting or telework was coined y Nilles 975, 994 to descrie a work arrangement in which employees work outside the conventional workplace e.g., home and interact with their managers and co-workers y way of computer-ased technology. There has een considerale research on telecommuting over the last 5 years, initially spawned y the oil emargo of the early mid-970s and the desire to sustitute communications for transportation Nilles, Carlson, Gray and Hanneman 976. Telecommuting has een linked to overall societal enefits such as reduced traffic congestion, reduced transport costs and decreased real-estate costs see Bailey and Kurland 00, and enefits to workers that participate in telework programs such as higher jo satisfaction and decreased interference etween work and family life Igaria and Guimaraes 999; Duxury, Higgins, and Newfeld 998. Recent developments in information technology IT have accelerated the trend toward telecommuting, expanding the range of firms and industries where telecommuting is viale. Currently, more than 6.5 million people in the US telework at least one day per month of their normal work schedule International Telework Association 000.

2 While considerale work has een done examining on the implications of telecommuting for workers, significantly less research has examined the managerial decision to support telework programs and how telework arrangements should e designed Bailey and Kurland 00. Recent studies of telecommuting arrangements have found that managerial support is critical for oth the existence and success of telework programs Powell and Mainiero 999; Anderson, Coffey and Byerly 00; Mannering and Mokhtarian 995; Mokhtarian and Salomon 997. However, studies have suggested that support for these programs is far from overwhelming with a survey of 4000 uropean managers reporting limited interest Huws, Korte and Roinson 990, p. 73 and other research showing that managers have significant concerns aout the difficulties in monitoring employee ehavior remotely Kurland and gan 999; Olson 98; Hamlin 995 and the potential for telecommuting to compromise jo performance Pearlson and Saunders 00. Formal research on telecommuting from the firm s perspective has focused on descriptive case studies or statistical studies involving correlations etween specific telework practices and outcomes Belanger and Collins 998. While these studies have helped understand telecommuting in practice, many researchers have noted that these studies have een hampered y the lack of an appropriate theoretical foundation upon which to ase hypotheses Bailey and Kurland 00; Hartman et al. 99; McCloskey and Igaria 998; Raghuram, Wiesenfeld and Garud 996. This has limited the ailities of these studies to gain insights into the process y which telecommuting causes various organizational outcomes and limited managerial prescriptions to studies of oserved est practice. In this paper we present a formal economic model of telecommuting focusing specifically on two types of managerial decisions identified in the previous literature: Which workers should e permitted to participate in telework programs? How should workers involved in telecommuting e monitored and compensated? While there are a variety of arrangements for working outside the conventional workplace Davenport and Pearlson 998, to facilitate modeling and empirical testing we focus specifically on telecommuting where workers with fixed offices occasionally work at home. This is y far the dominant means of telecommuting, does not have some of the more serious issues isolation, difficulty of mentoring associated with purely virtual firms or some of the other alternative work arrangements Cooper and Kurland 00, and is a relatively well defined decision that is presumaly faced y most managers. Using contract theory, we derive a set of relationships among manager s information aout

3 worker type, worker s individual characteristics, worker selection criteria for telecommuting programs, the intensity of incentives, and the alance etween sujective e.g., managerial review and ojective e.g., piece rates incentive systems. We then test these relationship using the 998 Workplace mployment Relationship Survey WRS98, a large survey dataset that captures oth workers and managers perspectives on telework arrangements. Overall, we find oth empirical and theoretical support for the argument that managers will e more likely to support telework arrangements when they are etter informed aout workers characteristics and that telecommuting employees should and do receive a higher proportion of sujectively determined incentive compensation in their overall pay. II. Model II. A. Modeling Approach and Description The theory of contracts is a widely used collection of approaches for examining interactions etween a small numer of economic agents see descriptions in Salanie 997 or Hart 989. The most common model within contract theory is the principal-agent model, which in its most stylized form involves a setting in which a principal e.g., a manager seeks to elicit costly effort from an agent e.g., a worker, ut cannot measure effort directly. As a result of this imperfect measurement, the principal must typically rely on various means to align the interests of the principal and the agents such as incentive contracts, monitoring or selection mechanisms. The telecommuting decision prolem discussed in the introduction naturally lends itself to a principal-agent formulation. In our model, a manager is facing a decision as to whether to allow a worker to work at home. The manager, however, faces two uncertainties. First, she may not e sure of the worker s propensity to engage in non-work activities during work time if allowed to work at home these may e due to personality, family situation or other factors which may e only partially oservale to the principal. Second, she may not e ale to oserve how much effort the agent exerts on work-related tasks. Clearly, the principal can etter monitor effort and does not have to contend with alternative non-work activities y her employees when they work in the office. However, disallowing telecommuting may not e optimal as it can e less costly to In the description of the model, we will follow the convention of referring to manager as she and worker as he. 3

4 the agent either explicitly or implicitly to accomplish the same amount of work with some telecommuting. The manager has two instruments that she can use to otain the optimal level of effort from her employees. First, she can choose which workers to trust to telecommute. Workers with a low inherent propensity to engage in outside activities can e allowed to work at home with a relatively low cost in terms of lost productivity. Thus, the manager faces a selection prolem in deciding which workers should e permitted to telecommute. Second, the manager can provide incentives to workers to telecommute so that they are more likely to engage in work-related activities. The efficacy of these incentives depends on the information the manager has aout performance and therefore must e considered jointly with the issue of monitoring and oservaility of effort. If effort is highly oservale either directly or through measurale surrogates such as worker output the principal can contract specifically on these oservale outcomes e.g., x% per sale made for a salesperson. However, if the manager can perceive worker effort, even if she cannot measure it explicitly, she can utilize sujective incentives, such as a onus ased on a managerial review. Thus, the manager faces a decision prolem as to how strong incentives should e and the optimal alance etween sujective and ojective incentives. Our model draws on insights from the multi-task principal agent model e.g., Holmstrom and Milgrom 99 which captures how incentives alter the division of effort etween work and nonwork tasks modified to account for varying agent characteristics. We also utilize an approach ased on the Baker, Gions and Murphy 994 model of sujective measurement in determining the mixture of sujective versus ojective incentives. II. B. Model Formulation We model the interaction etween a principal P and an agent A in which there is a decision as to whether the agent should e allowed to work at home. If the agent works at home, he can receive additional private enefits of working at home with a monetary value designated y. Workers have different types which correspond to how much time they must devote to capturing if allowed to work at home. A worker of type [, ] will spend t = / to Note that as with most principal-agent models with voluntary agent participation, the ojective is to maximize overall value since the agents overall utility is governed y their outside alternatives. We are not considering the issue of whether telework can e used to redistriute value etween the managers and the workers only the issue of whether more value can e created y such an arrangement. 4

5 capture private enefits if allowed to work at home high values of correspond to good types of workers. In addition, workers may have different productivity levels independent of their time allocation when working in the office as compared to working at home. Let p e the productivity of working in the office and p the corresponding productivity of working at home we make no assumptions aout which is larger. 3 If the agent spends time t on the jo, he creates value of p t for the principal working in the office, and p t working at home. The agent ears a private cost given y C t + t where C 0 < 0, C ' > 0, C ' ' > 0. Note that these assumptions imply that there is some positive numer t *, wherec t* = 0, and thus agents will exert some effort for a fixed wage. We also assume that is not so small so that that the agent works voluntarily and thus must earn a reservation utility of u 0. / < t * and The principal faces a decision prolem as to whether to allow the agent to work at home and therefore capture his private enefits, and what type of compensation structure to offer the agent to maximize her profit. We first consider two enchmark models, one in which the principal is constrained to pay a fixed wage a case where effort is totally unoservale, and the other where the principal can oserve effort completely and offer an effort time ased contract. Then we study the case where the worker s performance at home can only e partially oserved. Finally, we consider whether including sujective incentives can improve performance over the aove case. Case Unoservale Agent Action/Fixed-wage Contract. Without telecommuting, the principal solves the following optimization program: Max w π p t w, where w is the fixed wage. = s.t. IR w Ct u 0 IC t = argmax w C t t This is the standard formulation of the principal agent prolem where a principal maximizes her profit suject to the agent s individual rationality or participation constraint IR and the agent s incentive compatiility IC constraint, which ensures the agent to take the action the 3 Several researchers show that workers report that their productivity is higher at home Bailey and Kurland, 00, Belanger and Collins 998, although managers dout this Pearlson and Saunders 00. 5

6 principal intended when he is free to choose any action. Given a fixed wage the agent will always exert the no incentive effort t = t * to satisfy the IC condition. To satisfy the IR constraint, the principal will offer a wage w = u + C * and therefore receive a payoff of: 0 t π = p t u C * * 0 t If telecommuting is allowed, the principal must account for different types of agents. From incentive compatiility, the agent offered a wage w will exert effort t to satisfy: Max w C t + t + t s.t. t t,t Since we assume that if the agent works at home, he always otain his outside enefit, t = t *. The principal then must solve a program similar to the one efore accounting for the productivity difference of working at home and the agent s modified IR constraint which includes his outside enefit: P: Max w π p t w = s.t. IR w C t + t + t u IC t = t * 0 Then we have P s maximal payoff: π = pt * u0 C t* + p If the principal can oserve then she will offer telecommuting when it is more profitale that is, π > π. Comparing equation and we get we get p. p p t * p * = will e the threshold that P should offer the option of telecommuting. p p t * From the aove results we have the following proposition: 6

7 Proposition: Given either p or p fixed, * increases with p p under some mild constraints 4 ; This means the higher the productivity in office, the less likely the principal will offer telecommuting to her workers. On the other hand, the higher the productivity at home, the more likely P will offer telecommuting. Proposition is consistent with the oserved telecommuting trends. As IT increases the relative productivity of working at home, through improved information processing and data transfer capaility, there will e a greater numer of workers engaging in telecommuting. Case Fully Oservale ffort/time-ased Contracting. To simplify the prolem and remove the issue of productivity differences, we will assume going forward that productivity is the same at home and at work, thus p = p = p. If effort t is oservale, then the agent can implement the optimal action using a linear incentive contract where w= s+ at s is the ase salary, and a is the rate for the agent s time. Without telecommuting, the agent solves: Max s + at Ct to t determine his optimal effort. The first order condition yields: C ' t = a which we will rewrite as t = T a [the conditions on C guarantee it can e inverted to otain T ]. Since at optimum the agent s IR constraint always inds, the principal s prolem then ecomes Maxπ = pt a C T a u a 0, which yields an optimal solution for the principal of π = pt p C T p u0. As efore, with telecommuting the agent s prolem ecomes slightly more complex: Max s + at C t + + with a solution t = T a. After imposing t the agent s IR constraint, the principal s prolem ecomes Maxπ = p T a C T a + u0 which has a solution a * = p, the same as in the notelecommuting example. Thus π π * = p a. Case 3 Partial Oservaility/Ojective Incentive Contracting. Suppose now that the agents actions are fully oservale in the office, ut only partially oservale at home. Following Baker et al. 994, we assume that if the worker spends time t on his jo, he generates an p 4 t * When p is fixed, there is no constraints; When p is fixed, we need enough. to e positive. That is, is large 7

8 ojective performance measurement t, where is a random distortion factor with =. The worker can oserve, ut the manager can only oserve t. Thus, the worker will decide his optimal working time according to the realization of, giving rise to a moral hazard prolem. We also extend our definition of type to include the variance in measured output; we assume a good worker has a lower Var than a ad worker and therefore will e less inclined to act opportunistically. For tractaility we assume the worker s cost function is C t = γ t m, where m is a positive numer. Using this cost function, the solution to the no-telecommuting analysis in Case earlier p yields a * = p andπ = u0 + m. We now consider the solution when the agent works at 4γ home. Because the agent oserves, he will choose the est response according to the realization of. The agent s prolem then ecomes: solution of a t* =. γ Max s t + at γ t + + m + with a The manager s prolem is to choose the optimal a to maximize her expected payoff where is the expectation operator: Max pt γ t m u a { } From the first order condition we get * p a = which yields p p π = + + m u 4γ 0. Comparing the profit in this scenario with telecommuting and no telecommuting yields p π > π => * = p 4γ This result has a numer of implications. First, when comparing the partial oservaility case to the full oservaility case, the manager is more reluctant to allow telecommuting when effort is imperfectly oservale and thus requires workers to e of higher type efore allowing telecommuting. Second, since = + Var, increased variance of and thus the less 8

9 precise the ojective measurement, the higher the * and fewer workers will e allowed to work at home. These results are consistent with arguments in previous research that to successfully adopt telecommuting, firms should try to make the output more structured and easier to measure that is, they should structure work to increase the precision of ojective measurement. From the aove, we can easily show that if two workers are the same, the one who has a lower Var will e more likely to get the option of working at home. Similarly, if two workers Var are the same, the one who has higher is more likely to reach the threshold of *. This suggests our first hypothesis: H: Workers with high or low Var are more likely to have the privilege of working at home. The information of and Var are valuale to the manager, in that it enales managers to give the privilege of working home to the right person. Without this information, the manager will face a more difficult decision prolem and thus may e more inclined to limit telecommuting. In fact, in the Appendix we show that in the asence of this information, even the manager knows the average is aove the threshold *, she will not offer the telecommuting option to her workers. From the aove we draw our second hypothesis: H: Managers who know more aout their workers are more likely to support telecommuting. One way in which managers learn aout the characteristics of their workers is through formal communications policies as well as informal interaction. Therefore we would expect that: Ha: Workers with more communication with managers are more likely to get the option of working at home. Case 4 Partial Oservaility/Sujective and Ojective Incentive Contracting. In this section we assume that esides the ojective measurement, oth managers and workers can oserve a sujective measurement of effort. In this case we analyze whether this sujective measurement can e used to reach a more efficient outcome. 9

10 Following Baker et al. 994, we assume that this sujective measurement takes the form ofε t, where ε is a positive random variale with ε =. In this way we allow the sujective measurement to e imperfect if Var ε > 0. The worker can oserve ε, which means the worker may e ale to exploit suspected iases in the manager s evaluation technology. The manager can only seeε t and will use this measure to offer a onus ofεt β, where β is the onus rate. Because εt is the sujective perception of the manager, she can not put this in an explicit contract and this is the difference etween ε t and t. However, under certain conditions when oth sides are patient enough, which means discount rate is small enough, or the manager can commit to award the onus through some mechanism such as reputation see Baker et al. 994, contracting ased on this sujective measurement can e feasile. We assume that the manager can credily commit to pay a onus ased on her sujective performance assessment. When working at home, the worker s prolem ecomes: has an optimal solution of t = Max s + ta + εtβ C t + which t a + εβ. The principal s prolem is to set the hourly rate a γ and onus rate β. Intuitively, we can see that if Var = 0 perfect ojective measures then the manager would set a = p and β = 0 to achieve the optimal outcome with no sujective incentive component. Similarly, ifvar β = 0, she will set a = 0 and β = p to use only the sujective measurement. If neither of them is perfect, the manager will use oth and achieve a etter result, which is shown elow. The principal s prolem is That is, to maximize Max a,β pt C t + + u0, ε a + εβ a + εβ p γ + + m u γ γ, ε p[ ε ] p[ ] F.O.C. a = and β = * ε. ε * 0 This yields a maximum expected payoff of: 0

11 [ ] [ ][ ] [ ] [ ] [ ] [ ] 0 4 u m p p p p p p = ε γ ε ε ε ε γ ε ε γ π Comparing the results of Case 3 only ojective measures and this case we find that profits are increased y [ ] [ ] 4 ε γ p y using sujective incentives. Because Var + =, and ε ε Var + =, this quantity is always positive, unless 0 = Var. Thus, when ojective and sujective measures are oth imperfect, the principal can do etter y using oth, and thus more likely to offer telecommuting to her workers. This yields our final hypothesis: H3: Managers who use more sujective evaluations are more likely to support telecommuting. III. Data and mpirical Results III. A. Data Introduction We use data from the 998 Workplace mployment Relationship Survey WRS98 to test our hypotheses. The WRS98 dataset is a collection of surveys performed y personnel from a stratified sample of over 000 British estalishments with 0 or more employees Cully et al WRS98 comprises three sources of data, all collected etween Octoer 997 and June 998: interviews with,9 managers with day-to-day responsiilities for personnel matters in the estalishment; 947 interviews with worker representatives from a recognized trade union or consultative committee for the estalishment; and 8,37 completed questionnaires from a sample of employees working at each participating estalishment. For clarity, all references to specific variales are written in italics and we keep their initial variale names from WRS98. Please refer to Tale for variale definitions, description and coding. II. A. i. Manager survey workplace/estalishment level There are three key constructs needed to test our hypotheses. First, the dependent variale in our analysis is the option of telecommuting which is explicitly asked in the survey.

12 Second, we need some measure of how well managers can assess the type of worker as per our theoretical model. While there are no direct measures of worker type in the survey there are two general categories of activities that would enale a manager to assess worker type: information gathering and communications. For information gathering, the survey asks whether any of the following 9 records is kept for this estalishment, like Laor Costs, Productivity, Quality of product or service, Laor turnover, Asenteeism and Workforce training. We create a variale KRCPR_SUM which is the count of all kinds of records kept for the estalishment to measure the information endowment of the manager. The manager is also asked the ways the quality of work is monitored at this workplace. Variale KHOWMON_SUM is generated as the count of ways of monitoring. Our assumption is that the more ways in monitoring, the etter information the manager would have aout the work quality of her workers. For communications, the survey includes measures on the use of quality circles DCIRCLS, whether the firm has a system of riefings for workers DBRIF and whether there are joint consultative committees including managers and workers DJOINT. The manager is also asked whether she regularly gives employees or their representatives any information aout the following items such as internal investment plans, the financial position of the estalishment and staffing plans. This can partly reflect the manager s attitude to information sharing and communication these constructs are captured y variales DINVPLAN, DFINANC, and DSTAFFIN -- see Tale for definitions. Overall, although the measurements are far from perfect, comined as a group, they do seem to provide an ojective measurement of how much information the manager has aout her workers. We therefore factor analyze these measures and construct a composite variale F_INFO as proxy for the manager s information endowment for details, see Tale. The lower the value of F_INFO means the etter the manager knows her workers. Finally, we require some measures of the use of sujective evaluation. In the survey, the manager is asked what proportion of non-managerial employees at the workplace have their performance formally appraised, which is measured y variale FMASPR. We use this variale as proxy of sujective evaluation. We also add several control variales. It has een argued that teamwork may adversely influence the manager s attitude to telecommuting. We include CTAM, proportion of workers work in

13 formally designated teams. The structural nature of the jo may also e a factor that influences the manager s decision on telecommuting. Managers are asked main methods y which employees are made aware of their jo responsiilities. We use the count of these methods, CJOBDNW_SUM, to control for this effect. We also include additional controls for type of activities in the workplace KACTIVI, the industry ASIC, whether it is a cost or profit center KRFIC, estalishment size LOG_ZALLFT and gender composition ZTOT_RATIO of the workplace. These are all variales that commonly appear in empirical studies of strategic human resource practices. Upon data completion, the oservations we use in the regression range from 5 to 775 due to missing data on some variales. III. A. ii. mployee Survey individual level The second part of our data is an employee survey in the WRS98 data. In each workplace a random sample of 5 workers are selected. If there are fewer than 5 employees, then all workers are sent a questionnaire. Workers are asked a similar question aout whether working at home is availale to them, which is coded in variale B34. To test H, we need measurements of the type of worker. One advantage of the employee survey is that workers are asked to put a score of -6 on the following questions: I share many of the value of my organization and I feel loyal to my organization. We elieve the corresponding variales B5A and B5D are good measurements of the employee s attitude to his jo, and thus a good proxy for the type of the worker. Workers are also asked how their managers communicate with them on issues like future plans, staffing, changes to work practice, payment issues, and health, which is captured y variales B7A-B7. We use these variales as proxies of the communication effect in manager s information endowment. We also use education D5 and occupation D9 as control variales for individual employeelevel tests. Because these measurements are not availale at workplace level in the manager survey, we use the sample average of education and occupation, DU_AVG and OCU_AVG, as corresponding measurements at workplace level. Other variales that we use as controls are gender D and age D, and the unique numer SRNO assigned to each estalish in the survey data this implicitly controls for all workplace-specific characteristics that were included in the manager survey analysis. Descriptive statistics of our data can e found in Tale. 3

14 III. B. conometric Models and Results Hypotheses H and Ha are from the worker s perspective and tested at employee individual level, using data from employee survey. H and H3 are from manager s perspective and tested at workplace level, using data from the manager survey. Since we are predicting a 0/ variale, we employ logit model in our regressions. Specifically, H and Ha are tested at employee level using the following model: LnPr/ Pr = β + β X + β X + Controls + ε 0 type Communication where Pr is the proaility of having the option of working at home. and B5D, which reflect worker type; X Communication X type includes variales B5A includes variales B7A-B7, which measure communication with managers. Control variales are education D5, occupation D9, gender D and age D. SRNO is used to control the heterogeneity among workplaces. H and H3 are tested at manager level using the following model: Ln Pr/ Pr = β 0 + β * F _ INFO + β * FMASPR + Controls + ε where Pr is the proaility of offering telecommuting to workers. H corresponds to the sign and significance of β, and H3 corresponds to β. As detailed in II.A, controls include industry, the degree of structure of the jo, and sample average of education and occupation DU_AVG and OCU_AVG at the working place. Other controls are industry, size and gender composition. III. B. i. mployee Data: Testing H and Ha We first test hypotheses H and Ha using data from the employee survey. We limit our analysis to employees from a random suset of aout 800 estalishments due to constraints on the size of non-linear fixed effects models in our regression software results are similar with other random susets. The regression results are reported in Tale 3. Standardized errors are also reported. In Column we test only the effects of worker s type to telecommuting. Two variales that depict worker s type, B5A the degree of sharing the value of the organization and B5D the degree of feeling loyal to the organization, are oth significant at % level. Due to the coding of the scale, the negative coefficients imply that employees who report eing more committed to the organization are more likely to telecommute. 4

15 We then add variales of communication with manager to the regression B7A-B7. The effects of discussing future plans B7A, changes to work practice B7C and payment issues B7D are significant at % to 0% level. The other two variales, discussion of staffing issues and health/safety issues are not significant. The results are quite reasonale, ecause the former three variales are closely related to working at home, while the later two are not. In this way, Ha is supported y the data. It is possile that employees in a higher skill occupation or with a higher education are more loyal to the organization, and communicate more with the manager. So our results might e driven y occupation and education. To test this, we add them in the regression and results are reported in Column 3. They are oth significant at the p<.0 level. The change in the coefficient of B5A is not sustantial, and it remains significant at p<.0. B5D is not significant in Columns and 3. This appears to e due to colinearity etween B5A and B5D the correlation coefficient etween them is and they are jointly significant. If we drop B5A and run the logit regression again B5D is significant at % level, while coefficients of other variales remain unchanged. Overall, the aove regression results provide strong empirical support on H and Ha. Although we can not rule out other possiilities, we elieve that the worker s type and communication with the manager do influence whether he is allowed to work at home. III. B. ii. Manager Data: Testing H and H3 Now we use data from the manager survey to test the effects of information endowment and sujective measurement on managers telecommuting decisions. First, we test only the effect of information endowment. Results from our aseline regression are listed in Column of Tale 4. As expected, the information endowment factor is significant at a 5% level. Because of the way this factor is constructed, the oserved negative coefficient is consistent with our expectation. When we add the sujective measurement variale FMASPR Column, Tale 4, this variale is significant at % level, while the coefficient of F_INFO remains almost unchanged from -.54 to -.68 and still significant. FMASPR is coded such that a low value means more use of sujective measurement, so the negative coefficient of FMASPR means that sujective measurement is positively correlated to managers willingness of offering telecommuting. 5

16 These analyses support our argument in H and H3. Other interesting findings are: unionization and the proportion of female workers at the estalishment are also positively correlated with telecommuting. One simple explanation of our results is that information intensive workplaces or occupations are more likely to e sujectively measured, have high information sharing and utilize telecommuting. To determine whether our results are driven y this alternative explanation, we repeat our analyses controlling for education DU_AVG and occupation OCU_AVG. We find that education is not significant possily ecause of the correlation with occupation, ut occupation has a significant effect on telecommuting. It shows that if a workplace is mainly consists of knowledge workers and administrative employees, it is more likely that the manager will offer telecommuting. One explanation is that information work can e done more easily at home than physical work. However, even in this modified specification, the main results continue to hold, consistent with H and H3. IV Conclusion and Limitations Overall, ased on our theoretical modeling, we conclude that managers will e more likely to offer telecommuting to workers that are less likely to shirk at home and those for which they have good information aout their propensity to shirk. In addition, we show that sujective incentives can often complement ojective measures, especially when ojective measures are more difficult to utilize or unavailale for home work settings. Using WRS98 data we find support for these hypotheses at the worker and workplace manager level. Managers with etter information aout their workers allow more telecommuting and use a higher proportion of sujective incentives. Workers that are more committed to the firm are more likely to e ale to telecommute. We elieve this work complements existing research on telecommuting y extending the results to two critical managerial prolems in telecommuting worker selection, incentives, developing a formal model of telework, and testing this model using empirical data. While the data is not perfect, especially since we do not oserve worker motivations for engaging in telework and cannot directly oserve worker type, and our model clearly astracts away some important issues, such as how telework comines with office work, we elieve our analysis adds new insights into how firms should manage their workers when telecommuting is a viale option. 6

17 Acknowledgments: We would like to thank the seminar participants at Wharton School of University of Pennsylvania, the associate editor, and one anonymous referee for the helpful comments. We acknowledge the UK Department of Trade and Industry, the conomic and Social Research Council, the Advisory, Conciliation and Aritration Service and the Policy Studies Institute as the originators of the WRS 98 data, and The UK Data Archive as its distriutor. This material is ased upon work supported y the National Science Foundation under Grant No. IIS Any remaining errors are, of course, the sole responsiility of the authors. Reference Anderson, S.., Coffey, B.S., and Byerly, R.T. Formal Organizational Initiatives and Informal Workplace Practices: Links to Work-Family Conflict and Jo-related Outcomes, Journal of Management 8:6, 00, pp Bailey, D.. and Kurland, N.B. A Review of Telework Research: Findings, New Directions, and Lessons for the Study of Modern Work, Journal of Organizational Behavior 3:4, 00, pp Baker, G., Gions, R., and Murphy, K.J. "Sujective Performance Measures in Optimal Incentive Contracts," The Quarterly Journal of conomics 09:4, 994, pp Belanger, F., and Collins, R.W. Distriuted Work Arrangements: A Research Framework, Information Society 4:, 998, pp Cooper, C.D. and Kurland, N.B. Telecommuting, Professional Isolation, and mployee Development in Pulic and Private Organizations, Journal of Organizational Behavior 3, 00, pp Cully, M., Woodland, S., O Reilly, A., and Dix, G. Britain at Work: as Depicted y the 998 Workplace mployee Relations Survey, London: Routledge, 999. Davenport, T.H., and Pearlson, K. Two Cheers for the Virtual Office, Sloan Management Review 39:4, 998, pp. 5. Duxury, L., Higgins, C., and Newfeld, D. Telework and the alance etween work and family: Is telework part of the prolem or part of the solution? In Igaria M, Tan, M ds., The Virtual Workplace, Hershey, PA: Idea Group. 998, pp Hamlin, H. mployee s Perspectives on One Dimension of Laour Flexiility: working at a distance, Work, mployment & Society 9, 995, pp Hart, O. An conomist s Perspective on the Theory of the Firm, Columia Law Review 89:7, 989, pp Hartman R.I., Stoner C.R., and Arora, R. Developing Successful Organizational Telecommuting Arrangements: Worker Perceptions and Managerial Prescriptions, S.A.M. Advanced Management Journal 57:3, pp Holmstrom, Bengt., and Milgrom, P. "Multitask Principal-Agent Analyses: Incentive Contracts, Asset Ownership, and Jo Design," Journal of Law, conomics and Organization 7, 99, pp.4-5. Huws, U., Korte, W., and Roinson, S. Telework: Towards the lusive Office. Chichester, UK: Wiley, 990. Igaria, M., and Guimaraes, T. xploring Differences in mployee Turnover Intentions and its Determinants among Telecommuters and Non Telecommuters, Journal of Management Information Systems 6:, 999, pp

18 International Telework Association Kurland, N.B., and gan, T.D. Telecommuting: Justice and Control in the Virtual Organization, Organization Science 0:4, 999, pp Mannering, J.S., and Mokhtarian, P.L. Modeling the Choice of Telecommuting Frequency in California: an xploratory Analysis, Technological Forecasting and Social Change , pp Mokhtarian, P.L., and Salomon I. Modeling the Desire to Telecommuting: the Importance of Attitudinal Factors in Behavioral Models, Transportation Research 3:, 997, pp McCloskey, D.W., and Igaria, M. A review of the empirical research on telecommuting and directions for future research, In Igaria M, Tan, M ds., The Virtual Workplace. Hershey, PA: Idea Group. 998, pp Nilles, J.M. Telecommunications and Organizational Decentralization, I Transactions On Communications 3 975, pp Nilles, J., Carlson, F. R., Gray, P., and Hanneman, G. The Telecommunications-Transportation Ttradeoff: Options for Tomorrow. New York: John Wiley, 976. Nilles J.M. Making Telecommuting Happen. Van Nostrand Reinhold: New York, 994. Pearlson, K.., and Saunders, C.S. There's No Place Like Home: Managing Telecommuting Paradoxes, Academy of Management xecutive 5:, 00, pp.7-8. Olson, M. H. New Information Technology and Organizational Culture, Management Information Systems Quarterly 4:4, 98, pp Powell, G.N., and Mainiero, L.A. Managerial Decision Making Regarding Alternative Work Arrangements, Journal of Occupational and Organizational Psychology 7:, 999, pp Raghuram, S., Wiesenfeld, B., and Garud, R. Distance and Propinquity: A New Way to Conceptualize Work, Proceedings of the Telecommuting 96 Conference, Jacksanville, FL, 996. Salanie, B. The conomics of Contracts, The MIT Press, 997. Appendix For simplicity, we use fixed wage contract presented in Model and assume that p = p = p. This yields a threshold of * = p. We now show that even when the manager knows that her workers average type equals to p, that is, = p, the manager may still do not want to offer the option of telecommuting to her workers. For simplicity, suppose that is uniformly l l distriuted at [ p, p + ], with a density. From model, we get π π = p, so if l the manager offers telecommuting to all of her workers, her expected payoff would e: + l p + λ π π = l p d. Let l = λp, then we have π π = ln. p l λ λ + λ Thus if ln >, the manager will not offer telecommuting to her workers, even half of her λ λ worker qualifies for this policy. And we see that this always holds as long as λ 0,. Tale : Variale Definition and Descriptive Statistics Note: coding is as per WRS98 source data except as noted y * From the manager survey: 8

19 Variale name definition Coding Mean SD IFAMILY_WORKHOM whether non-managerial employees are allowed to work at home in normal working hours yes =; no = KRCPR_SUM* Count of types of workplace records Count KHOWMON_SUM* Count of types of monitoring Count DCIRCLS Whether have specific prolem groups yes =; no = DBRIF Whether have deriefing system yes =; no = DJOINT Whether there are joint consultative committees, yes =; no = works councils, etc DINVPLAN Whether manager regularly gives out information yes =; no = DFINANC aout investment, finance and staffing issues yes =; no = DSTAFFIN yes =; no = FMASPR The portion of performance appraisal to workers Rank of-6, most, 6-very few ANYMP Any worker Unionized in the workplace yes =, no = Please refer to section III for the description of other control variales: CTAM, CJOBDNW_SUM, ASIC, KACTIVI, KRFIC, LOG_ZALLFT and ZTOT_RATIO. From the employee survey: Variale name definition Coding Mean SD B34 Whether working in or from home in normal yes =; no = hours is availale to you B5A I sharing the value of the organization -6, means strongest B5D I feel loyal to my organization -6, means strongest How often do workers communicate with the manager aout - B7A Future plans for the workplace -4. means frequently B7B Staffing issues 4 means hardly B7C Changes to work practices B7D Pay issues B7 Health and safety at work Other variales DU_AVG* The sample mean of education in this workplace -6 The igger, the more education. xcept 6 OCU_AVG* The sample mean of occupation in this workplace -9, The smaller, the more administrative and research occupation SRNO Workplace identifier Please refer to section III for the description of other control variales: D, D, D5, D9. Tale : Construction of F_INFO: Factor Analysis Summary.a: igenvalues : Factor loadings of the eight IT measures KNOWMON KRCPR DCIRCLS DBRIF DJOINT DINVPLAN DFINANC DSTAFFIN Note: Conventional criteria applied to the results in Tale a suggest that the first factor is the only non-noise factor. Tale 3 Testing for H and Ha Logit regression dependent variale: B34 Whether working in or from home in normal hours is availale to the employee Independent Variales Col Col Col3 Col4 9

20 D ***.6*** D9 -.47*** -.453*** B5A -.6*** -.6*** -.478*** B5D -.09** *** B7A -.09*** -.8** -.58*** B7B B7C -.68*** -.05* -.0* B7D.96***.4***.30** B Other Controls Workplace Dummies, Gender, Age Workplace Dummies, Gender, Age Workplace Dummies, Gender, Age Workplace Dummies, Gender, Age N Pseudo R * -p<., ** -p<.05, *** -p<.0, standard error in parentheses Tale 4: Testing H and H3 Logit regression dependent variale: IFAMILY_WORKHOM whether non-managerial employees are allowed to work at home in normal working hours Independent Variales Col Col Col 3 F_INFO -.54** -.68** -.33** FMASPR -.4*** -.6*** ZTOT_RATIO.649*.890**.45** LOG_ZALLFT DU_AVG OCU_AVG *** *** *** *** Industry Dummies and Major Activity Dummies, Team, Jo Structure ANYMP.393**.437** Industry Dummies and Industry Dummies and Other Controls Major Activity Major Activity Dummies, Team, Jo Dummies, Team, Jo Structure Structure N Pseudo R * -p<., ** -p<.05, *** -p<.0, standard error in parentheses 0