Abstract. BACKGROUND: Proponents of telework arrangements assert that those who telework have more control over their...

Transcribed Image Text:

Abstract. BACKGROUND: Proponents of telework arrangements assert that those who telework have more control over their work and family domains than their counterparts who are not permitted to work from home. OBJECTIVE: Using Karasek's theory we hypothesized that the relationship between demands (hours in work per week; hours in childcare per week) and strain (work role overload; family role overload) would be moderated by the number of hours the employee spent per week teleworking (control). METHODS: To determine how the number of telework hours relates to work role overload and family role overload, we follow the test for moderation and mediation using hierarchical multiple regression analysis as outlined by Frazier et al. [50] We used survey data collected from 1,806 male and female professional employees who spent at least one hour per week working from home during regular hours (i.e. teleworking). RESULTS: As hypothesized, the number of hours in telework per week negatively moderated the relation between work demands (total hours in paid employment per week) and work strain (work role overload). Contrary to our hypothesis, the number of hours in telework per week only partially mediated the relation between family demands (hours a week in childcare) and family role. overload (strain). CONCLUSIONS: The findings from this study support the idea that the control offered by telework is domain specific (helps employees meet demands at work but not at home). Keywords: Telecommute, Karasek's demand-control, moderation, mediation (two thirds of the respondents had personal incomes of $60,000 or more per year). More than half the respon- dents were "knowledge workers" with just over 60% of the sample working in managerial and professional po- sitions. In the majority of families represented in this sample (75%) both partners work for pay outside the home. Two thirds of the respondents have children at home. Complete details on the study and the respon- dents can be found in Duxbury and Higgins [50]. The sample used in this study was selected from this larger data set as follows: all respondents were parents who worked full time and spent at least one hour tele- working in the week prior to their completion of the survey. We examined the literature to determine what is currently known with respect to the type of employ- ees who typically telework. This review determined that researchers have looked at two different types of teleworkers: employees in management and profes- sional positions and women in clerical positions [23]. Throughout the remainder of this paper we focus on the former group of teleworkers (i.e., managers and professionals), as findings on this group speak to con- cerns employees have with respect to the recruitment and retention of employees in this group in modern organizations [51]. One thousand eight hundred and nine (1809) respondents fit this criterion. Three obser- vations were removed as they were outliers reported teleworking more hours than they reported working in total. This left 1806 responses for analysis. Sample demographics are shown in Table 1. The majority of the teleworkers in the sample were female (60.3% female). Virtually all respondents were middle aged (average age of 45.1 years) married (80.5%) indi- viduals in the "full nest stage of the lifecycle. Parents spent an average of 11.7 hours per week in childcare. Respondents spent an average of 40.7 hours per week in paid employment and 11.0 hours per week in tele- work (26% of their total hours). The fact that respon- dents had been with their current organization for 11.5 years and in the current job for 5.5 years suggests that experienced professionals are more likely to be permit- ted to telework than are their more junior colleagues. The men in the sample differed from the women in a number of ways (see Table 1). The women in the sample were younger than the men and had spent fewer years working for their current employer. Also noteworthy are the data showing that the women in the sample spent more hours in childcare per week than their male colleagues and reported higher levels of work role overload and family role overload. The men in the sample, on the other hand, spend more hours per week in paid employment. These gender differences are consistent with those commonly reported in the lit- erature. There were no gender differences in time spent in telework per week and work role overload. 3.2. Procedure To conduct our analysis we tested for moderation and mediation using hierarchal regression analysis. The following outlines the procedure used for both tests. The test for moderation shows whether the mod- erator variable significantly influences the outcome variable beyond that of the predictor variable. To do this, the predictor and moderator variables should be centered and standardized to reduce problems asso- ciated with multicollinearity (i.e., high correlations) among the variables in the regression equation [56, 58]. A product term is used to establish the effect of the moderator variable beyond that of the predictor. To form the product term, we multiplied together the pre- dictor and moderator variables using the newly cen- tered and standardized terms [57]. If the product term is significant after controlling for predictor and moder- ator variables, than moderation exists. We conduct this analysis using hierarchal regression analysis. Three steps are used as outlined by Fraizer et al. [50]. Demographic variables are included in the first step to control for the influence of potential of con- founding variables. Predictor and moderator variables are included in the second step to account for individ- ual effects. Product terms are included in the final step to see whether the product term significantly accounts for variance beyond that of the individual contributions of predictor and moderator variables. To further interpret the moderating variable, many scholars chose to show the influence of the moderating variable through visual means. To do so one may fol- low the procedure recommended by Cohen et al. [55] and create two simple linear regressions of the pre- dictor variable (i.e., work role overload on hours in work per week) and plot the results. The plot shows two points along the X-axis: high (+1SD) and low (- 1SD) values of the predictor variable. It also shows two regression lines: high (+1SD) and low (-1SD) values of the moderating variable (i.e., hours in telework per week). If the test for moderation proved unsuccessful, it is possible that the telework variable is actually a me- diating variable. Testing for mediation explores the idea that the predictor variable influences a mediation variable which in turn influences an outcome variable. ences in tenure (operationalized as years in organiza- tion and years in job) because some researchers have suggested employees use telework arrangements as a way to "escape" organizational stress that has devel- oped over time [23]. Finally, our decision to control for time in eldercare per week is consistent with our decision to define family demands in terms of hours per week in childcare and the fact that relatively few respondents (25%) spent time each week in eldercare activities. Work Role Overload 3.4 3.3 3.2 3.1 3 2.9 2.8 2.7 2.6 2.5 Low Telework Hours High Telework Hours 4. Results To determine how the number of telework hours re- lates to work role overload and family role overload, we follow the test for moderation and mediation using hierarchical multiple regression analysis as outlined by Frazier et al. [55]. To compare our data to previous research, Table 2 presents means, standard deviations, and correlations for each of the measures. The signifi- cant correlation between work role overload and fam- ily role overload is consistent with what has been re- ported in the literature [1,2,4] and supports the idea that spillover occurs between work and family domains (i.e. people who are overwhelmed at work are likely to "bring it home" and vice versa). In support of Hypoth- esis 1 and 3, we find strong positive associations be- tween predictor and outcome variables within the work environment (i.e. hours in work per week and work role overload; B 0.247, p < 0.01) and the family envi- ronment (i.e. hours in childcare per week and family role overload; B = 0.321, p < 0.01). Strong support for these two hypotheses supports our decision to test the ability of hours per week in telework to moderate both of these relationships. = 4.1. Moderation We followed the procedures outlined by Frazier et al. [55] to test hypotheses 2 and 4 (i.e. number of hours in telework per week moderates the relation- ship between work/family demands and work/family overload). As shown in Table 3, Hypothesis 3, which posited that hours spent in telework per week (i.e. con- trol) would moderate the relationship between work demands (hours in work per week) and work role over- load, was supported (B = -0.040, p < 0.05). The R change associated with the interaction term was 0.003 (p < 0.05). - 1 SD (30.21 hrs) +1 SD (51.25 hrs) Employed Hours Fig. 1. The moderating role of hours in telework per week on the hours in work per week - work role overload relationship. Mean employed hours = 40.73. = To further interpret the interaction effect, we plot predicted values for the outcome variable (work role overload) for representative groups. As shown in Fig. 1, low hours in telework per week was associated with both the lowest levels of work role overload (Y = 2.75 when hours in work per week was low) and the high- est levels of work role overload (Y: 3.3 when hours in work per week was high). When hours per week in telework are high, on the other hand, work role over- load increases at a slower rate (i.e. values of Y = 2.85 for low hours in work per week and Y = 3.2 for high hours in work per week). These finding indicate that the control over work demands offered by telework be- come more important as hours spent in work week increase. per We also used hierarchical regression analysis used to test for a possible curvilinear relationship within our model. We tested for curvilinearity by computing a squared telework hour term and entering it in after the control and demographic variables. In the regression model, a curvilinear relationship is evident if the addi- tion of this quadratic term results in significant incre- mental variance after the linear effect has been taken into account [56]. As shown in Table 3, there was no significance in the squared term coefficient ( = 0, n.s.) and the model did not achieve a better fit with addition of this variable. This additional analysis pro- vides additional support for the linear model tested in this study. 1 An Excel file created to calculate these predicted values is avail- able from Linda Duxbury. Table 1 Gender differences Male Female Total Variable N M SD N M SD N M SD Age 712 46.71 10.62 1082 44.08 9.58 1794 45.12* 10.08 Years with org 715 13.63 50.00 1087 10.21 8.88 1802 11.57* 32.02 Years in job 714 6.37 6.75 1088 5.00 5.10 1802 5.54** 5.85 Eldercare hours 718 1.08 3.06 1088 1.93 6.20 1806 1.59** 5.21 Employed hours 717 43.03 11.25 1087 39.20 9.72 1804 40.73** 10.52 Telework hours 718 11.09 13.50 1088 11.00 12.39 1806 11.03 12.84 Childcare hours 718 8.64 12.22 1088 13.63 19.88 1806 11.65** 17.42 Work role overload 718 2.87 0.84 1088 3.13 0.88 1806 3.03 0.87 Family role overload 670 2.70 0.87 1018 3.07 0.92 1688 2.92* 0.92 p <0.01. Note. Significant gender differences are shown in the table: *p < 0.05; **, Again, one may follow the procedures outlined by Fra- zier et al. [55]. The mediation hypothesis states that (i) the mediating variable will significantly relate to the outcome variable, (ii) the predictor variable will signif- icantly relate to the mediating variable, and (iii) the re- lation between the predictor variable and the outcome variable will be smaller when controlling for the me- diating variable. Each of these three tests will be con- ducted independently using hierarchal regression anal- ysis. If these three steps are satisfied with significant regression coefficients than partial mediation occurs. To test for complete mediation, the effect of the pre- dictor variable on the outcome variable when control- ling for the mediating variable should be zero. Given our sample size (N = 1688), we had sufficient power to detect medium to large mediation effects [59]. 3.3. Measures = We used Caplan et al.'s [52] seven item measure to quantify work role overload. Respondents were asked to respond, using a five point Likert Scale (1 = Seldom, 5 Almost Always) to questions such as: "How often do your colleagues make too many demands on you?" and "How often do the number of tasks you have to do at work exceed the amount of time you have to do them?" We averaged the scores to each of the seven line items to form a single value. Higher scores reflect greater work role overload. The Cronbach's alpha for this sample was 0.92. We used a nine item measure which combined items from Caplan et al. [52] and Bohen and Viveros- Long [7] overload measures to quantify family role overload. Respondents were asked to respond, using a five point Likert Scale (1 = Seldom, 5 = Almost Al- ways) to questions such as "How often do your chil- dren make too many demands of you?" and "How of- ten do the number of tasks you have to do at home ex- ceed the amount of time you have to do them? We av- eraged the scores to attain a single value for family role overload, such that higher scores reflect greater fam- ily role overload. The Cronbach's alpha for this sample was 0.91. To quantify the number of hours spent per week in telework (i.e. control) we asked respondents the fol- lowing question: "At present, approximately how many hours per week do you spend doing job-related work at home during regular office hours?" Responses ranged from 1 to 60 hr/week. The mean was 11.03 and the standard deviation 12.84. Work demands were quantified by asking respon- dents the following question: "At present, approxi- mately how many hours per week do you spend in paid employment overall? Responses ranged from 1 to 85 hr/week. The mean was 40.73 and standard devi- ation was 10.52. Family demands were operationalized by asking re- spondents the following question: "At present, approx- imately how many hours per week do you spend in child-care or activities with your children? Just over half the respondents (57% of the sample) reported that they spent at least an hour per week in childcare. The mean of the entire sample was 11.65 and standard devi- ation 17.42 while the mean and standard deviation for just those who provide childcare was 20.44 and 18.79, respectively. = = To mitigate the possibility of spurious results we controlled for age, gender (male 1, female 2), tenure (in years) and hours per week in eldercare activ- ities during the analysis. We controlled for age because it generally approximates career-life stage, which may play a role in the development of coping mecha- nisms [53]. We controlled for gender because previ- ous works suggest that women report higher levels of role overload than men [54] a gender difference that is apparent in our data set. We controlled for differ- 98 L. Duxbury and M. Halinski/An examination of the relationship between hours in telework and role overload Table 2 Descriptive statistics and correlations Variable M 1. Sex 2. Age SD 1.60 0.49 45.12 10.09 -0.128** 1 2 3 4 5 6 7 8 9 3. Yrs with org 2.05 0.88 -0.052* 0.172** 4. Yrs in current job 5.54 5.85 -0.115** 0.375** 0.501** 5. Eldercare hours 1.59 5.21 0.080** 0.132** 0.030 0.091** 6. Employed hours 40.73 10.52 -0.178** 0.016 -0.001 0.050* 0.027 7. Telework hours 11.03 12.84 -0.003 0.029 -0.014 0.028 0.044 8. Childcare hours 11.65 17.42 0.140** -0.210** -0.039 9. Work Overload 3.03 0.87 0.141** -0.030 0.019 10. Family Overload 2.92 0.92 0.201** -0.172** -0.044 0.034 -0.063* 0.199** -0.064** -0.025 -0.090** 0.129** 0.247** 0.057* 0.099** -0.028 -0.061* -0.046* 0.002 0.321** 0.343** Note. N = 1806. *p < 0.05; **p < 0.01. Table 3 Testing moderator effects using hierarchal multiple regressions Model 1: Work role overload Model 2: Family role overload Step and variable Employee demographics Age B AR 0.026** B AR 0.075** -0.039 -0.155** Gender 0.13** 0.17** Years with org 0.101 0.153 Years in job 0.049* 0.005 Eldercare hours 0.089** Control variables 0.076** 0.077** Employed hours 0.243** Telework hours 0.004 -0.045* Childcare hours 0.251** Quadratic variable 0.0 0.001 Telework hours 0.0 0.0 Interaction effects 0.003* 0.0 Employed hrs X Telework hrs -0.040* Childcare hrs X Telework hrs 0.021 Quadratic Interaction Effect 0.0 0.0 Employed hr X Telework hr 0.0 Childcare hrs X Telework hrs 0.0 Note. N = 1806; CI = Confidence Interval; *p < 0.05; **p < 0.01. We followed the same procedure to test Hypothe- sis 4 (i.e. the number of hours in telework per week moderates the relationship between hours per week in childcare and family role overload). The data did not support Hypothesis 4 (see Table 3) as the interaction term (3 = 0.021, n.s.) was not significant. 4.2. Mediation Following the suggestion of Frazier et al. [55] we undertook follow-up analysis to determine if the hours per week in telework mediated the relationship be- tween hours in childcare per week and family role overload. Table 4 shows the analyses necessary to ex- amine the meditation hypothesis as described above. In this second analysis, the variables of interest are hours Following the steps outlined above (Step 1), we found childcare hours (predictor) significantly related to fam- ily role overload (outcome; = 0.017, p < 0.01). To establish the relation between hours in childcare per week (outcome) and hours in telework per week (me- diator), we regressed hours in childcare on hours in telework (Step 2). While this relationship was signifi- cant (B = -0.034, p < 0.05) it was not in the antici- pated direction. Rather, the data indicate that contrary to our expectations hours per week in telework is nega- tively related to hours per week in childcare. This find- ing contradicts an idea that is popular in the work-life literature parents will choose to telework to better accommodate their family circumstances [23]. Finally, to test whether hours per week in telework was related to family role overload, we regressed family role over- 1 - work more so that their flexible schedule [63] can ac- commodate the increased demands of their parent re- sponsibilities [23]. Our results show the opposite, sug- gesting that the greater the demands at home (i.e. the higher the responsibility for childcare) the less likely the employee is to telework. Again, while it is dif- ficult to establish the direction of causality here, we can speculate why this might be the case. First, it may be that parents with greater responsibilities at home chose to segment their roles as a way to increase their work efficiency (i.e. get more done "at the office" than when working from home). Alternatively, segmenta- tion theory would suggest that by keeping the work and family domains separate, employees can recover from within-domain strains more effectively [64]. What we do not know, however, is whether or not such a separa- tion allows parents to recover from childcare demands at work, or from heavy work demands at home! Fi- nally, it may be the case that employers are less likely to facilitate telework arrangements for employees with younger children/higher family demands than for those with fewer demands under the assumption that tele- work arrangements are not a substitute for childcare. In any case, these findings are unfortunate given our data showing that people with higher demands at home (i.e. those more likely to benefit from the use of this work arrangement) are less likely to telework. Future work in the area is needed to determine why this might be the case. 6. Conclusions The findings from this study are relevant to employ- ers, teleworkers and researchers. From the employer perspective, these findings support the idea that em- ployers who want to help employees cope with heavy work demands and reduce the deleterious impacts of work role overload within their organizations should encourage their professional employees to spend some of their work time teleworking. In particular, our find- ing that telework influences the reduction of work role overload for those who spend a higher number of hours in work per week, suggest that managers should en- courage and facilitate teleworking opportunities for employees who spend more than 45 hours per week in work. They should also encourage (not discourage) telework for people who have higher demands at home as telework mediates the relationship between hours in childcare per week and family role overload. The data from this study is also of relevance to em- ployees who are feeling overwhelmed by all the work that they have to do in that it suggests that such em- ployees will benefit from the adoption of telework ar- rangement. While the benefits will largely be felt at work (reduction of work role overload) they may also experience a slight improvement of their circumstances at home (slight reduction of family role overload). Finally, the results are of benefit to researchers in a number of ways. First, researchers of Karasek's de- mand control model frequently limit its conceptualiza- tion to that of the work domain [66], and do not ad- dress the potential influence of the family domain on the work domain. Our results fill this gap by providing further validation of Karasek's demand-control model of strain and suggesting a number of ways that this model can be updated to reflect today's workforce (i.e., including family domain in analysis). Second, our re- sults provide empirical evidence that reinforce the idea that telework can be considered to be a form of con- trol in that it negatively moderates and mediates the re- lation between demands and strains. Third, the results provide evidence that telework arrangements can ben- efit employees and employers alike and offer sugges- tions on why that might be the case. Fourth, the results suggest that researchers looking at the work-life con- flict construct and what contributes to role strain need to include domain specific role overload in their stud- ies. Finally, this study furthers our understanding of what contributes to work and family role overload and suggests mechanisms through which employers and employees alike can cope with increasing demands at work and at home. This paper is not without its limitations. First, the data presented here is self-reported. Secondly, partic- ipants in this study are from larger organizations. Fu- ture studies should focus on undertaking this type of study in small and medium sized organizations where demands and flexibility may be quite different. Third, responses from multiple organizations were used to as- semble the data set. The fact that we combined multi- ple types of organizations (i.e. public, private, not for profit) while a strength of this study, might introduce a number of confounds into the analysis. Future re- searchers could remove this limitation by focusing on a single type of organization. Fourth, while we con- trolled for age, gender, years in organization, years in job, and time in eldercare per week there are other variables such as management level [44] that should be considered in future work in this area. Finally, our conceptualization of family demands is limited to that of childcare demands (i.e., parenting). While this deci- sion is consistent with the advice of Hausser and col- leagues [40], it is a rather narrow definition of fam- ily demands. Future research in this area could bene- fit from a more comprehensive measure of family de- mands (i.e. time per week in childcare, eldercare, home maintenance) or a different operationalization of this construct (i.e. responsibility for dependent care for ex- ample, rather than time in role). In summary, despite these limitations, we feel that this study does add to our understanding of the tele- work arrangements in several important ways. First, this research is grounded in theory. Second, the fact that the sample is large and many potential confounds are controlled for during the analysis gives us confi- dence in the generalizability of our findings. Third, the fact that we conceptualized telework as a continuous variable allowed us to examine the impact of hours in telework per week on various types of strain. Such an approach provides more insight than conceptualizing telework as a categorical variable (no telework, part time telework, fulltime telework). [11] Coverman S. Role overload, role conflict, and stress: Address- ing consequences of multiple role demands. Social Forces. 1989; 67; 965-982. [12] Arora R. Are Americans really abject workaholics? Gallup Poll Briefing. 2004; 1-4. [13] Bond J, Thompson C, Galinsky E, Prottas D. Highlights of the National Study of the Changing Workforce, Number Three. New York: Families and Work Institute, 2002. [14] Kodz J, Davis S, Lain D, Sheppard E, Rick J, Strebler M, Bates P, Cummings J, Meager N, Anxo D, Gineste S, Trinczek R. Working Long Hours in the U.K.: A Review of the Re- search Literature, Analysis of Survey Data and Cross-national Organization Case Studies. Employment Relations Research Series No. 16. London: Department of Trade and Industry, 2003. [15] Organization for Economic Co-operation and Development (OECD). Clocking In and Clocking Out: Recent Trends in Working Hours. Policy brief. [cited 2004]. Retrieved from: http://www.oecd.org/dataoecd/42/49/33821328.pdf. [16] Duxbury L, Higgins C. Work-life conflict in Canada in the new millennium: A status report. Ottawa: Health Canada, 2003. [17] Higgins C, Duxbury L, Johnson K. Exploring the link be- tween work-life conflict and demands on Canada's health care system (Report Three). Ottawa: Health Canada, 2004. [18] Barnett R C, Baruch G K. Women's involvement in multiple their work, the number of hours they work, and the lo- cation where they work, which affects their commuting time and total time away from home [39, p. 3]. These authors state that schedule control (control over when and where one works) complements the concept of job control (control over how work is done) as put forward by Karasek [24,25]. A substantial body of research exists that supports the idea that the more control an employee has over their work time, the greater their ability to balance work and family demands [39]. No studies could be found, however, linking schedule control to role over- load. This study addresses this gap in the literature by examining whether or not schedule control (oper- ationalized as the number of hours an employee tele- works per week) moderates: (i) the relationship be- tween work demands and work role overload and (ii) the relationship between family demands and family role overload. Our review of the extant literature provides both the- oretical and empirical justification for hypothesizing time in telework as a moderator of the relationship be- tween work demands and work role overload. In terms of theory, Karasek's buffering hypothesis [24,28] leads us to expect that telework arrangements will make it easier for employees with high job demands to man- age all they are trying to do [39]. Also relevant to our argument is the idea that telework allows employees to work more efficiently and get more done in a shorter period of time. Tietze and Musson [22] suggest sev- eral mechanisms by which this might occur including increased autonomy over the structure of the working day, fewer distractions from work, and a greater ability to maximize work outputs. The literature also provides support for our con- tention that time in telework per week will mod- erate the relationship between family demands and family role overload by making it easier for em- ployees to manage time at home. A number of re- searchers [45-47] make a link between telework ar- rangements, greater control over the work-life inter- face and an enhanced ability to fulfill household re- sponsibilities and optimize time management. Other researchers [48,49] expect that employees' families will benefit from the increased ability teleworkers have to control the pacing and scheduling of their work by making it easier for them to schedule their family time. In terms of theory, Karasek's buffering hypothesis [24, 28] leads us to expect that telework arrangements will make it easier for employees with high family demands to manage all they are trying to do [39]. A recent review of the Demand-Control model by Hausser et al. [40] suggests that Karasek's framework is ideally suited for a study such as this one. First, Hausser et al. [40] identified 83 studies published be- tween 1998 and 2007, which suggests that Karasek's framework is still very applicable in today's environ- ment and well accepted throughout academia. Second, Hausser et al.'s [40] collection of studies highlights 35 studies with emotional exhaustion as the outcome variable, 14 studies with psychological distress and 15 studies with depression. The fact that all three of these variables are well known consequences of role over- load [2,5,9,16,18,19] positions Karasek's framework capable of examining role overload. Finally, Hausser et al. [40] concludes the demand-control model is ca- pable of describing the relationship between domain specific variables (i.e., work role overload) and cross domain variables (i.e., hours teleworking), as long as sufficient thought is put into matching demands with controls [40]. 2.4. Hypothesis The above review of the literature supports the fol- lowing hypothesis: Hypothesis 1: The number of hours spent in paid employment per week (i.e. work demands) is posi- tively related to work role overload. Hypothesis 2: Time in telework will moderate the relationship between work demands and work role overload. Hypothesis 3: The number of hours spent in child- care per week (i.e. family demands) is positively related to family role overload. Hypothesis 4: Time in telework will moderate the relationship between family demands and family role overload. 3. Methodology 3.1. Sample In 2011-12, 25,021 Canadian employees working for 87 medium to large public, private and not-for- profit sector employers (i.e., 100 or more employ- ees) responded to a national survey exploring issues associated with work-life balance. Respondents were very well educated (22% with a college diploma, 38% with one university degree, 17% with at least one post graduate degree) and socio-economically advantaged Abstract. BACKGROUND: Proponents of telework arrangements assert that those who telework have more control over their work and family domains than their counterparts who are not permitted to work from home. OBJECTIVE: Using Karasek's theory we hypothesized that the relationship between demands (hours in work per week; hours in childcare per week) and strain (work role overload; family role overload) would be moderated by the number of hours the employee spent per week teleworking (control). METHODS: To determine how the number of telework hours relates to work role overload and family role overload, we follow the test for moderation and mediation using hierarchical multiple regression analysis as outlined by Frazier et al. [50] We used survey data collected from 1,806 male and female professional employees who spent at least one hour per week working from home during regular hours (i.e. teleworking). RESULTS: As hypothesized, the number of hours in telework per week negatively moderated the relation between work demands (total hours in paid employment per week) and work strain (work role overload). Contrary to our hypothesis, the number of hours in telework per week only partially mediated the relation between family demands (hours a week in childcare) and family role. overload (strain). CONCLUSIONS: The findings from this study support the idea that the control offered by telework is domain specific (helps employees meet demands at work but not at home). Keywords: Telecommute, Karasek's demand-control, moderation, mediation (two thirds of the respondents had personal incomes of $60,000 or more per year). More than half the respon- dents were "knowledge workers" with just over 60% of the sample working in managerial and professional po- sitions. In the majority of families represented in this sample (75%) both partners work for pay outside the home. Two thirds of the respondents have children at home. Complete details on the study and the respon- dents can be found in Duxbury and Higgins [50]. The sample used in this study was selected from this larger data set as follows: all respondents were parents who worked full time and spent at least one hour tele- working in the week prior to their completion of the survey. We examined the literature to determine what is currently known with respect to the type of employ- ees who typically telework. This review determined that researchers have looked at two different types of teleworkers: employees in management and profes- sional positions and women in clerical positions [23]. Throughout the remainder of this paper we focus on the former group of teleworkers (i.e., managers and professionals), as findings on this group speak to con- cerns employees have with respect to the recruitment and retention of employees in this group in modern organizations [51]. One thousand eight hundred and nine (1809) respondents fit this criterion. Three obser- vations were removed as they were outliers reported teleworking more hours than they reported working in total. This left 1806 responses for analysis. Sample demographics are shown in Table 1. The majority of the teleworkers in the sample were female (60.3% female). Virtually all respondents were middle aged (average age of 45.1 years) married (80.5%) indi- viduals in the "full nest stage of the lifecycle. Parents spent an average of 11.7 hours per week in childcare. Respondents spent an average of 40.7 hours per week in paid employment and 11.0 hours per week in tele- work (26% of their total hours). The fact that respon- dents had been with their current organization for 11.5 years and in the current job for 5.5 years suggests that experienced professionals are more likely to be permit- ted to telework than are their more junior colleagues. The men in the sample differed from the women in a number of ways (see Table 1). The women in the sample were younger than the men and had spent fewer years working for their current employer. Also noteworthy are the data showing that the women in the sample spent more hours in childcare per week than their male colleagues and reported higher levels of work role overload and family role overload. The men in the sample, on the other hand, spend more hours per week in paid employment. These gender differences are consistent with those commonly reported in the lit- erature. There were no gender differences in time spent in telework per week and work role overload. 3.2. Procedure To conduct our analysis we tested for moderation and mediation using hierarchal regression analysis. The following outlines the procedure used for both tests. The test for moderation shows whether the mod- erator variable significantly influences the outcome variable beyond that of the predictor variable. To do this, the predictor and moderator variables should be centered and standardized to reduce problems asso- ciated with multicollinearity (i.e., high correlations) among the variables in the regression equation [56, 58]. A product term is used to establish the effect of the moderator variable beyond that of the predictor. To form the product term, we multiplied together the pre- dictor and moderator variables using the newly cen- tered and standardized terms [57]. If the product term is significant after controlling for predictor and moder- ator variables, than moderation exists. We conduct this analysis using hierarchal regression analysis. Three steps are used as outlined by Fraizer et al. [50]. Demographic variables are included in the first step to control for the influence of potential of con- founding variables. Predictor and moderator variables are included in the second step to account for individ- ual effects. Product terms are included in the final step to see whether the product term significantly accounts for variance beyond that of the individual contributions of predictor and moderator variables. To further interpret the moderating variable, many scholars chose to show the influence of the moderating variable through visual means. To do so one may fol- low the procedure recommended by Cohen et al. [55] and create two simple linear regressions of the pre- dictor variable (i.e., work role overload on hours in work per week) and plot the results. The plot shows two points along the X-axis: high (+1SD) and low (- 1SD) values of the predictor variable. It also shows two regression lines: high (+1SD) and low (-1SD) values of the moderating variable (i.e., hours in telework per week). If the test for moderation proved unsuccessful, it is possible that the telework variable is actually a me- diating variable. Testing for mediation explores the idea that the predictor variable influences a mediation variable which in turn influences an outcome variable. ences in tenure (operationalized as years in organiza- tion and years in job) because some researchers have suggested employees use telework arrangements as a way to "escape" organizational stress that has devel- oped over time [23]. Finally, our decision to control for time in eldercare per week is consistent with our decision to define family demands in terms of hours per week in childcare and the fact that relatively few respondents (25%) spent time each week in eldercare activities. Work Role Overload 3.4 3.3 3.2 3.1 3 2.9 2.8 2.7 2.6 2.5 Low Telework Hours High Telework Hours 4. Results To determine how the number of telework hours re- lates to work role overload and family role overload, we follow the test for moderation and mediation using hierarchical multiple regression analysis as outlined by Frazier et al. [55]. To compare our data to previous research, Table 2 presents means, standard deviations, and correlations for each of the measures. The signifi- cant correlation between work role overload and fam- ily role overload is consistent with what has been re- ported in the literature [1,2,4] and supports the idea that spillover occurs between work and family domains (i.e. people who are overwhelmed at work are likely to "bring it home" and vice versa). In support of Hypoth- esis 1 and 3, we find strong positive associations be- tween predictor and outcome variables within the work environment (i.e. hours in work per week and work role overload; B 0.247, p < 0.01) and the family envi- ronment (i.e. hours in childcare per week and family role overload; B = 0.321, p < 0.01). Strong support for these two hypotheses supports our decision to test the ability of hours per week in telework to moderate both of these relationships. = 4.1. Moderation We followed the procedures outlined by Frazier et al. [55] to test hypotheses 2 and 4 (i.e. number of hours in telework per week moderates the relation- ship between work/family demands and work/family overload). As shown in Table 3, Hypothesis 3, which posited that hours spent in telework per week (i.e. con- trol) would moderate the relationship between work demands (hours in work per week) and work role over- load, was supported (B = -0.040, p < 0.05). The R change associated with the interaction term was 0.003 (p < 0.05). - 1 SD (30.21 hrs) +1 SD (51.25 hrs) Employed Hours Fig. 1. The moderating role of hours in telework per week on the hours in work per week - work role overload relationship. Mean employed hours = 40.73. = To further interpret the interaction effect, we plot predicted values for the outcome variable (work role overload) for representative groups. As shown in Fig. 1, low hours in telework per week was associated with both the lowest levels of work role overload (Y = 2.75 when hours in work per week was low) and the high- est levels of work role overload (Y: 3.3 when hours in work per week was high). When hours per week in telework are high, on the other hand, work role over- load increases at a slower rate (i.e. values of Y = 2.85 for low hours in work per week and Y = 3.2 for high hours in work per week). These finding indicate that the control over work demands offered by telework be- come more important as hours spent in work week increase. per We also used hierarchical regression analysis used to test for a possible curvilinear relationship within our model. We tested for curvilinearity by computing a squared telework hour term and entering it in after the control and demographic variables. In the regression model, a curvilinear relationship is evident if the addi- tion of this quadratic term results in significant incre- mental variance after the linear effect has been taken into account [56]. As shown in Table 3, there was no significance in the squared term coefficient ( = 0, n.s.) and the model did not achieve a better fit with addition of this variable. This additional analysis pro- vides additional support for the linear model tested in this study. 1 An Excel file created to calculate these predicted values is avail- able from Linda Duxbury. Table 1 Gender differences Male Female Total Variable N M SD N M SD N M SD Age 712 46.71 10.62 1082 44.08 9.58 1794 45.12* 10.08 Years with org 715 13.63 50.00 1087 10.21 8.88 1802 11.57* 32.02 Years in job 714 6.37 6.75 1088 5.00 5.10 1802 5.54** 5.85 Eldercare hours 718 1.08 3.06 1088 1.93 6.20 1806 1.59** 5.21 Employed hours 717 43.03 11.25 1087 39.20 9.72 1804 40.73** 10.52 Telework hours 718 11.09 13.50 1088 11.00 12.39 1806 11.03 12.84 Childcare hours 718 8.64 12.22 1088 13.63 19.88 1806 11.65** 17.42 Work role overload 718 2.87 0.84 1088 3.13 0.88 1806 3.03 0.87 Family role overload 670 2.70 0.87 1018 3.07 0.92 1688 2.92* 0.92 p <0.01. Note. Significant gender differences are shown in the table: *p < 0.05; **, Again, one may follow the procedures outlined by Fra- zier et al. [55]. The mediation hypothesis states that (i) the mediating variable will significantly relate to the outcome variable, (ii) the predictor variable will signif- icantly relate to the mediating variable, and (iii) the re- lation between the predictor variable and the outcome variable will be smaller when controlling for the me- diating variable. Each of these three tests will be con- ducted independently using hierarchal regression anal- ysis. If these three steps are satisfied with significant regression coefficients than partial mediation occurs. To test for complete mediation, the effect of the pre- dictor variable on the outcome variable when control- ling for the mediating variable should be zero. Given our sample size (N = 1688), we had sufficient power to detect medium to large mediation effects [59]. 3.3. Measures = We used Caplan et al.'s [52] seven item measure to quantify work role overload. Respondents were asked to respond, using a five point Likert Scale (1 = Seldom, 5 Almost Always) to questions such as: "How often do your colleagues make too many demands on you?" and "How often do the number of tasks you have to do at work exceed the amount of time you have to do them?" We averaged the scores to each of the seven line items to form a single value. Higher scores reflect greater work role overload. The Cronbach's alpha for this sample was 0.92. We used a nine item measure which combined items from Caplan et al. [52] and Bohen and Viveros- Long [7] overload measures to quantify family role overload. Respondents were asked to respond, using a five point Likert Scale (1 = Seldom, 5 = Almost Al- ways) to questions such as "How often do your chil- dren make too many demands of you?" and "How of- ten do the number of tasks you have to do at home ex- ceed the amount of time you have to do them? We av- eraged the scores to attain a single value for family role overload, such that higher scores reflect greater fam- ily role overload. The Cronbach's alpha for this sample was 0.91. To quantify the number of hours spent per week in telework (i.e. control) we asked respondents the fol- lowing question: "At present, approximately how many hours per week do you spend doing job-related work at home during regular office hours?" Responses ranged from 1 to 60 hr/week. The mean was 11.03 and the standard deviation 12.84. Work demands were quantified by asking respon- dents the following question: "At present, approxi- mately how many hours per week do you spend in paid employment overall? Responses ranged from 1 to 85 hr/week. The mean was 40.73 and standard devi- ation was 10.52. Family demands were operationalized by asking re- spondents the following question: "At present, approx- imately how many hours per week do you spend in child-care or activities with your children? Just over half the respondents (57% of the sample) reported that they spent at least an hour per week in childcare. The mean of the entire sample was 11.65 and standard devi- ation 17.42 while the mean and standard deviation for just those who provide childcare was 20.44 and 18.79, respectively. = = To mitigate the possibility of spurious results we controlled for age, gender (male 1, female 2), tenure (in years) and hours per week in eldercare activ- ities during the analysis. We controlled for age because it generally approximates career-life stage, which may play a role in the development of coping mecha- nisms [53]. We controlled for gender because previ- ous works suggest that women report higher levels of role overload than men [54] a gender difference that is apparent in our data set. We controlled for differ- 98 L. Duxbury and M. Halinski/An examination of the relationship between hours in telework and role overload Table 2 Descriptive statistics and correlations Variable M 1. Sex 2. Age SD 1.60 0.49 45.12 10.09 -0.128** 1 2 3 4 5 6 7 8 9 3. Yrs with org 2.05 0.88 -0.052* 0.172** 4. Yrs in current job 5.54 5.85 -0.115** 0.375** 0.501** 5. Eldercare hours 1.59 5.21 0.080** 0.132** 0.030 0.091** 6. Employed hours 40.73 10.52 -0.178** 0.016 -0.001 0.050* 0.027 7. Telework hours 11.03 12.84 -0.003 0.029 -0.014 0.028 0.044 8. Childcare hours 11.65 17.42 0.140** -0.210** -0.039 9. Work Overload 3.03 0.87 0.141** -0.030 0.019 10. Family Overload 2.92 0.92 0.201** -0.172** -0.044 0.034 -0.063* 0.199** -0.064** -0.025 -0.090** 0.129** 0.247** 0.057* 0.099** -0.028 -0.061* -0.046* 0.002 0.321** 0.343** Note. N = 1806. *p < 0.05; **p < 0.01. Table 3 Testing moderator effects using hierarchal multiple regressions Model 1: Work role overload Model 2: Family role overload Step and variable Employee demographics Age B AR 0.026** B AR 0.075** -0.039 -0.155** Gender 0.13** 0.17** Years with org 0.101 0.153 Years in job 0.049* 0.005 Eldercare hours 0.089** Control variables 0.076** 0.077** Employed hours 0.243** Telework hours 0.004 -0.045* Childcare hours 0.251** Quadratic variable 0.0 0.001 Telework hours 0.0 0.0 Interaction effects 0.003* 0.0 Employed hrs X Telework hrs -0.040* Childcare hrs X Telework hrs 0.021 Quadratic Interaction Effect 0.0 0.0 Employed hr X Telework hr 0.0 Childcare hrs X Telework hrs 0.0 Note. N = 1806; CI = Confidence Interval; *p < 0.05; **p < 0.01. We followed the same procedure to test Hypothe- sis 4 (i.e. the number of hours in telework per week moderates the relationship between hours per week in childcare and family role overload). The data did not support Hypothesis 4 (see Table 3) as the interaction term (3 = 0.021, n.s.) was not significant. 4.2. Mediation Following the suggestion of Frazier et al. [55] we undertook follow-up analysis to determine if the hours per week in telework mediated the relationship be- tween hours in childcare per week and family role overload. Table 4 shows the analyses necessary to ex- amine the meditation hypothesis as described above. In this second analysis, the variables of interest are hours Following the steps outlined above (Step 1), we found childcare hours (predictor) significantly related to fam- ily role overload (outcome; = 0.017, p < 0.01). To establish the relation between hours in childcare per week (outcome) and hours in telework per week (me- diator), we regressed hours in childcare on hours in telework (Step 2). While this relationship was signifi- cant (B = -0.034, p < 0.05) it was not in the antici- pated direction. Rather, the data indicate that contrary to our expectations hours per week in telework is nega- tively related to hours per week in childcare. This find- ing contradicts an idea that is popular in the work-life literature parents will choose to telework to better accommodate their family circumstances [23]. Finally, to test whether hours per week in telework was related to family role overload, we regressed family role over- 1 - work more so that their flexible schedule [63] can ac- commodate the increased demands of their parent re- sponsibilities [23]. Our results show the opposite, sug- gesting that the greater the demands at home (i.e. the higher the responsibility for childcare) the less likely the employee is to telework. Again, while it is dif- ficult to establish the direction of causality here, we can speculate why this might be the case. First, it may be that parents with greater responsibilities at home chose to segment their roles as a way to increase their work efficiency (i.e. get more done "at the office" than when working from home). Alternatively, segmenta- tion theory would suggest that by keeping the work and family domains separate, employees can recover from within-domain strains more effectively [64]. What we do not know, however, is whether or not such a separa- tion allows parents to recover from childcare demands at work, or from heavy work demands at home! Fi- nally, it may be the case that employers are less likely to facilitate telework arrangements for employees with younger children/higher family demands than for those with fewer demands under the assumption that tele- work arrangements are not a substitute for childcare. In any case, these findings are unfortunate given our data showing that people with higher demands at home (i.e. those more likely to benefit from the use of this work arrangement) are less likely to telework. Future work in the area is needed to determine why this might be the case. 6. Conclusions The findings from this study are relevant to employ- ers, teleworkers and researchers. From the employer perspective, these findings support the idea that em- ployers who want to help employees cope with heavy work demands and reduce the deleterious impacts of work role overload within their organizations should encourage their professional employees to spend some of their work time teleworking. In particular, our find- ing that telework influences the reduction of work role overload for those who spend a higher number of hours in work per week, suggest that managers should en- courage and facilitate teleworking opportunities for employees who spend more than 45 hours per week in work. They should also encourage (not discourage) telework for people who have higher demands at home as telework mediates the relationship between hours in childcare per week and family role overload. The data from this study is also of relevance to em- ployees who are feeling overwhelmed by all the work that they have to do in that it suggests that such em- ployees will benefit from the adoption of telework ar- rangement. While the benefits will largely be felt at work (reduction of work role overload) they may also experience a slight improvement of their circumstances at home (slight reduction of family role overload). Finally, the results are of benefit to researchers in a number of ways. First, researchers of Karasek's de- mand control model frequently limit its conceptualiza- tion to that of the work domain [66], and do not ad- dress the potential influence of the family domain on the work domain. Our results fill this gap by providing further validation of Karasek's demand-control model of strain and suggesting a number of ways that this model can be updated to reflect today's workforce (i.e., including family domain in analysis). Second, our re- sults provide empirical evidence that reinforce the idea that telework can be considered to be a form of con- trol in that it negatively moderates and mediates the re- lation between demands and strains. Third, the results provide evidence that telework arrangements can ben- efit employees and employers alike and offer sugges- tions on why that might be the case. Fourth, the results suggest that researchers looking at the work-life con- flict construct and what contributes to role strain need to include domain specific role overload in their stud- ies. Finally, this study furthers our understanding of what contributes to work and family role overload and suggests mechanisms through which employers and employees alike can cope with increasing demands at work and at home. This paper is not without its limitations. First, the data presented here is self-reported. Secondly, partic- ipants in this study are from larger organizations. Fu- ture studies should focus on undertaking this type of study in small and medium sized organizations where demands and flexibility may be quite different. Third, responses from multiple organizations were used to as- semble the data set. The fact that we combined multi- ple types of organizations (i.e. public, private, not for profit) while a strength of this study, might introduce a number of confounds into the analysis. Future re- searchers could remove this limitation by focusing on a single type of organization. Fourth, while we con- trolled for age, gender, years in organization, years in job, and time in eldercare per week there are other variables such as management level [44] that should be considered in future work in this area. Finally, our conceptualization of family demands is limited to that of childcare demands (i.e., parenting). While this deci- sion is consistent with the advice of Hausser and col- leagues [40], it is a rather narrow definition of fam- ily demands. Future research in this area could bene- fit from a more comprehensive measure of family de- mands (i.e. time per week in childcare, eldercare, home maintenance) or a different operationalization of this construct (i.e. responsibility for dependent care for ex- ample, rather than time in role). In summary, despite these limitations, we feel that this study does add to our understanding of the tele- work arrangements in several important ways. First, this research is grounded in theory. Second, the fact that the sample is large and many potential confounds are controlled for during the analysis gives us confi- dence in the generalizability of our findings. Third, the fact that we conceptualized telework as a continuous variable allowed us to examine the impact of hours in telework per week on various types of strain. Such an approach provides more insight than conceptualizing telework as a categorical variable (no telework, part time telework, fulltime telework). [11] Coverman S. Role overload, role conflict, and stress: Address- ing consequences of multiple role demands. Social Forces. 1989; 67; 965-982. [12] Arora R. Are Americans really abject workaholics? Gallup Poll Briefing. 2004; 1-4. [13] Bond J, Thompson C, Galinsky E, Prottas D. Highlights of the National Study of the Changing Workforce, Number Three. New York: Families and Work Institute, 2002. [14] Kodz J, Davis S, Lain D, Sheppard E, Rick J, Strebler M, Bates P, Cummings J, Meager N, Anxo D, Gineste S, Trinczek R. Working Long Hours in the U.K.: A Review of the Re- search Literature, Analysis of Survey Data and Cross-national Organization Case Studies. Employment Relations Research Series No. 16. London: Department of Trade and Industry, 2003. [15] Organization for Economic Co-operation and Development (OECD). Clocking In and Clocking Out: Recent Trends in Working Hours. Policy brief. [cited 2004]. Retrieved from: http://www.oecd.org/dataoecd/42/49/33821328.pdf. [16] Duxbury L, Higgins C. Work-life conflict in Canada in the new millennium: A status report. Ottawa: Health Canada, 2003. [17] Higgins C, Duxbury L, Johnson K. Exploring the link be- tween work-life conflict and demands on Canada's health care system (Report Three). Ottawa: Health Canada, 2004. [18] Barnett R C, Baruch G K. Women's involvement in multiple their work, the number of hours they work, and the lo- cation where they work, which affects their commuting time and total time away from home [39, p. 3]. These authors state that schedule control (control over when and where one works) complements the concept of job control (control over how work is done) as put forward by Karasek [24,25]. A substantial body of research exists that supports the idea that the more control an employee has over their work time, the greater their ability to balance work and family demands [39]. No studies could be found, however, linking schedule control to role over- load. This study addresses this gap in the literature by examining whether or not schedule control (oper- ationalized as the number of hours an employee tele- works per week) moderates: (i) the relationship be- tween work demands and work role overload and (ii) the relationship between family demands and family role overload. Our review of the extant literature provides both the- oretical and empirical justification for hypothesizing time in telework as a moderator of the relationship be- tween work demands and work role overload. In terms of theory, Karasek's buffering hypothesis [24,28] leads us to expect that telework arrangements will make it easier for employees with high job demands to man- age all they are trying to do [39]. Also relevant to our argument is the idea that telework allows employees to work more efficiently and get more done in a shorter period of time. Tietze and Musson [22] suggest sev- eral mechanisms by which this might occur including increased autonomy over the structure of the working day, fewer distractions from work, and a greater ability to maximize work outputs. The literature also provides support for our con- tention that time in telework per week will mod- erate the relationship between family demands and family role overload by making it easier for em- ployees to manage time at home. A number of re- searchers [45-47] make a link between telework ar- rangements, greater control over the work-life inter- face and an enhanced ability to fulfill household re- sponsibilities and optimize time management. Other researchers [48,49] expect that employees' families will benefit from the increased ability teleworkers have to control the pacing and scheduling of their work by making it easier for them to schedule their family time. In terms of theory, Karasek's buffering hypothesis [24, 28] leads us to expect that telework arrangements will make it easier for employees with high family demands to manage all they are trying to do [39]. A recent review of the Demand-Control model by Hausser et al. [40] suggests that Karasek's framework is ideally suited for a study such as this one. First, Hausser et al. [40] identified 83 studies published be- tween 1998 and 2007, which suggests that Karasek's framework is still very applicable in today's environ- ment and well accepted throughout academia. Second, Hausser et al.'s [40] collection of studies highlights 35 studies with emotional exhaustion as the outcome variable, 14 studies with psychological distress and 15 studies with depression. The fact that all three of these variables are well known consequences of role over- load [2,5,9,16,18,19] positions Karasek's framework capable of examining role overload. Finally, Hausser et al. [40] concludes the demand-control model is ca- pable of describing the relationship between domain specific variables (i.e., work role overload) and cross domain variables (i.e., hours teleworking), as long as sufficient thought is put into matching demands with controls [40]. 2.4. Hypothesis The above review of the literature supports the fol- lowing hypothesis: Hypothesis 1: The number of hours spent in paid employment per week (i.e. work demands) is posi- tively related to work role overload. Hypothesis 2: Time in telework will moderate the relationship between work demands and work role overload. Hypothesis 3: The number of hours spent in child- care per week (i.e. family demands) is positively related to family role overload. Hypothesis 4: Time in telework will moderate the relationship between family demands and family role overload. 3. Methodology 3.1. Sample In 2011-12, 25,021 Canadian employees working for 87 medium to large public, private and not-for- profit sector employers (i.e., 100 or more employ- ees) responded to a national survey exploring issues associated with work-life balance. Respondents were very well educated (22% with a college diploma, 38% with one university degree, 17% with at least one post graduate degree) and socio-economically advantaged