5. MODEL DESCRIPTION

5.1 Logit Modeling

The binary logit formulation is one method used to operationalize random utility theory, which assumes that people always select the alternative with the highest utility, but that utility can not be measured perfectly. Rather, the utility of an alternative, Uin, can be separated into deterministic, Vin, and random components, ein, in other words, Uin = Vin + ein. The deterministic portion can then be assumed to be a linear combination of independent variables, Vin = b'xin, where b is a vector of parameters to be estimated. The binary logit model states that the probability of choosing any alternative i is equal to the probability that the utility of alternative i is greater than the utility of alternative j (where j is not equal to i). The mathematical formulation is given below:

Pn(i) = Pr[Uin >= Ujn] = exp[mb'xin] / (exp[mb'xin] + exp[mb'xjn])

where

Pn(i) - the probability of choosing i

i - any particular alternative

Uin - utility of alternative i

m - a scale parameter

The procedure used in this research to estimate the preference model uses both SPSS/PC+ and SST. The first computer program was used to conduct t-tests between the dependent variables (telecommuting preference) and independent variables (see Appendix C, Tables C-1 to C-3). Each test compared the means for the independent variable for the two groups of preferrers and non-preferrers. For those variables where the p-value was less than 0.4, the correlation coefficients were calculated. If two variables were found to be strongly correlated and both measured similar attributes, the one that was not as highly correlated with the dependent variable was removed from consideration (for example, distance and time to the regular workplace). Next, a binary logit model was estimated on the remaining variables using SST. Variables were removed from the logit model based on the significance of the coefficients; only those with t-statistics greater than 1.64, corresponding to a level of significance of 0.1, were retained. The final step was to enter stepwise into the model the variables that had p-values greater than 0.4 from the t-tests. The variables with significant coefficients were accepted into the model.

5.2 Variable Description

For the third model, the dependent variable was chosen so that the respondents were classified as either preferring to telecommute from a center or from home. As for the first two models, the preference was initially defined by part E, question 7, only this time, any amount of telecommuting was considered a valid indicator of preference. Then, if the preferred amount of telecommuting from a center was greater than the amount from home, the respondent was said to prefer the center, and if the reverse was true, the respondent prefers the home. For those who did not answer question 7 or had equivalent amounts for both alternatives, the responses to part E, question 2, the ideal distribution of time at the three workplaces (see Appendix A), were used to further discriminate the telecommuting location preference. Of the six respondents who failed to answer either one or the other part of question 7, four were classified as center-preferrers and two as home-preferrers. Of the 28 respondents who preferred equal amounts from both locations, 11 were reclassified to center preference, 12 were reclassified as home preference, and 5 were unable to be reclassified since they also choose identical percentages of time at both the center and home. The five respondents who strictly preferred either location and the one respondent who did not prefer any form of telecommuting were removed from consideration.

The 62 independent variables can be grouped into three categories: factor scores, work-related, and household-related variables (see Appendix C, Tables C-1 to C-3). The four factors for job satisfaction and the five factors for work attitudes at each of the three work locations are described in sections 4.3 and 4.4, respectively.

There are 24 work-related independent variables in the data set (see Appendix C, Table C-2). Four dummy variables categorize each respondent according to the type of job held: manager, professional/technical, administrative support, or sales/marketing. The first two fields are likely to be more telecommutable than the first two since more of their tasks can be done independently. Also, sales/marketing staff may prefer telecommuting in order to be more mobile (and closer to clients) while working. Another three dummy variables capture the effect of the length of time worked with the current supervisor, for the employer (both equal to one if greater than 2 years), and in the current occupation (equal to one if greater than 10 years). On the one hand, more experience at work may lead to greater confidence by the supervisor and more support for telecommuting, but, on the other hand, more experience may make the respondent more reluctant to change the work arrangement. Three dummy variables also measure the effects of different work schedules: part-time, flex-time, and previous telecommuting experience. Those who have already experimented with their work schedule may be amenable to telecommuting as another scheduling option. On the other hand, the adjustments already made to their schedule may be satisfactory, and therefore, they may be less likely to consider telecommuting. Two of the non-binary variables include the number of hours of overtime worked in two weeks, which may function as a measure of workaholism, and the rating of how important client demands are as a distraction to the respondent.

Additional work-related variables include five variables taken straight from the survey that record the percentage of time the respondent spends working independently, face-to-face, remotely, at a specific location, and while traveling, respectively. High percentages of independent and remote work may lead the respondent to prefer telecommuting since there would be more opportunities to work away from the office, while high percentages of face-to-face work, work at a specific-location, and work-related travel may lead the respondent to prefer not to telecommute. Three dummy variables take into account the need for phone services, office equipment, and office support services while telecommuting. The lack of any of these services may account for the preference to not telecommute. Finally, four dummy variables measure the suitability of the job and the support from the manager for both telecommuting from a center and from home. These dummy variables were constructed similarly to the dependent variables from part E, questions 4 and 5 (see Appendix A), with amounts of telecommuting of one to three days a month or greater considered to indicate job suitability or manager support.

The first five household-related variables (see Appendix C, Table C-3) are common sociodemographic information: age, household size, education, household income, and gender (all of which are measured by categories except for household size). Older employees are more likely to be resistant to change, so they are less likely to prefer telecommuting. Larger households may lead to more distractions for telecommuters, but the need to provide family care may also drive certain workers (especially women) to prefer telecommuting. Those who are more educated and have a higher income are likely to have the space and equipment to make telecommuting possible. Further, they may be more likely to hold professional or managerial positions with sufficient autonomy to allow telecommuting.

The next six binary variables are concerned with the presence of children or other dependents in the household. The first specifically measures the presence of other dependents and the next three record the presence of children who are less than 2 years, from 2 to 5 years, and from 6 to 15 years of age, respectively. Then, two variables cover the presence of any children less than 6 and any children less than 16. The age ranges reflect the fact that different ages of children require different levels of care and can cause different levels of distraction for the potential telecommuter. Additionally, the three "less than" variables are crossed with the gender of the respondent (a dummy variable equal to 1 if the respondent is female) since in most households women are responsible for child care.

The remaining variables deal mainly with transportation implications. Relocation, a dummy variable that records the desire to relocate or the recent relocation of the respondent's residence farther from the regular workplace, may lead the employee to prefer telecommuting to minimize the impact of a longer commute. Both the distance to the regular workplace and the time it takes to get there are used to measure the strength of the geographical barrier to commuting. Finally, the lack of a car for commuting as measured by the number of vehicles per licensed driver and the number of vehicles per workers in the household may motivate an employee to consider telecommuting as a work option.

5.3 Preference for Telecommuting Center

The work attitudes factors, two of the six significant variables in the model, have the most significant coefficients. The first of these, the personal benefits at the telecenter, drives one to prefer the center if one stands to derive personal benefits from working there. The coefficient of the second factor, work responsibility at the regular workplace, has a negative sign meaning that if respondents have been given autonomy at the main office, they are unlikely to prefer to telecommute from a center in order to gain that responsibility. The second work drive variable, the number of overtime hours worked in two weeks, is a significant predictor of center preference. Those who work more overtime are more likely to prefer center-based telecommuting, perhaps in order to get more work done.

Of the remaining three variables, one is family-related and the other two are job- related. It is unclear how the presence of children less than 6 years in the household influences one to prefer center-based telecommuting. This may be a drive for telecommuting at the center in order to be near children at home, a constraint to working at home due to household distractions, or combination of the two effects. The suitability of the job for telecommuting from a center is also an important factor in center preference. Although an unsuitable job is usually considered an external constraint (and therefore should not appear in a preference model), respondents may consider this requirement when making their decision on whether to prefer telecommuting from a center or not. The age of the respondent significantly predicts center preference, with older categories of age making one more likely to prefer the telecenter. This unexpected result may be caused by the selection of older, more experienced, and trusted employees as prospective telecenter users.

5.4 Preference for Home

The home preference model includes three factors as significant variables; one is an independence drive and the other two are work drives. The variable for personal benefits at the center is significant again in this model. Here, it has a strong negative loading, so those who gain little personal benefit from the center will likely prefer home- based telecommuting instead. From the job satisfaction factors, high job frustration is shown to drive respondents to prefer telecommuting from home. The work ethic at home factor is the third significant factor variable. This variable can be interpreted as a desire for home telecommuting if one can be motivated and work effectively from home.

The remaining variables are split between household and work variables. As in the center preference model, the presence of children is a significant variable in the preference for home-based telecommuting. However, in the second model the important variable is the presence of children younger than two years of age, that is, those who need the most care. A larger number of persons in the household leads one not to prefer home- based telecommuting which may be an indicator of household distractions for in-home working. Similarly, those who have spent a long time in their current occupation are unlikely to prefer home-based telecommuting (strong negative coefficient) because they are probably more resistant to change in the workplace and are more likely to be comfortable with their current work arrangement. Although an unsuitable job is considered a constraint to the choice for telecommuting, job suitability is once again an important element in the preference to telecommute, which is consistent with modeling results reported in Mokhtarian and Salomon (1994c) for telecommuting from home.

5.5 Preference for Center vs. Home

Unlike the first two models, the significant variables in the third model are factor variables except for one. As in the first two models, personal benefits from the center is again important to telecommuting preference. In this model, the coefficient, which is very large (43.4), signifies a preference for center if the respondent expects to gain personal benefits there or a preference for home if one does not. Work ethic at home is strongly negative pointing to a preference for home if the respondent has a good work ethic at that location. Work ethic at the regular workplace is also strongly negative towards the home preference; however, it is unclear why working well at the main office would make one favor home-based over center-based telecommuting. If the family factor is important, the respondent is likely to prefer home over the center possibly in order to provide better dependent care and manage household responsibilities. Finally, older respondents are more likely to prefer telecommuting from a center over telecommuting from home perhaps because the center is more like the familiar regular workplace.

5.6 Discussion of Final Models

It might be questioned whether the inequality of the sample sizes for the study and control groups affects the ability of the models to accurately predict telecommuting preference. The data set contains about twice as many prospective telecenter users as home-based and non-telecommuters combined. Discrete choice modeling theory generally holds that a logit model estimated on an unrepresentative sample will still yield consistent coefficient estimates, with the possible exception of the constant term (Ben-Akiva and Lerman 1985). In order to test this, a random sample of the center users-to-be was taken such that the sample sizes for both study and control groups were equal. As expected, the three final models re-estimated on this reduced set of data showed, in general, similar values for coefficients and for the rho-squared statistic (see Appendix C, Tables C-4 to C-6).

Note: C and H indicate that high values on the variable are favorable to the preference for center and home, respectively.

The study can be used to enumerate the factors that are important in the decision to adopt telecommuting (see Table 5-5). The factor score variables were clearly important in all three models, as were assorted workplace and family variables. The variable for personal benefits at the center appeared in all models and usually had large coefficients. The effect of this variable may be exaggerated somewhat by the character of the sample. Since a majority of the respondents are prospective telecenter users, the benefits of the center are likely overrated because of these respondents' anticipation to use the centers. Work ethic from home was a significant indicator for preference in two models explaining the tendency to prefer home for those who feel they can work well at home. The other variable to appear in two models, age of the respondent, showed a preference for the center for older age groups. Although the presence of children is a significant variable in two models, the variable is defined differently for each model and favors different preferences. The change from home to center preference for the older age group provides evidence for the effect of household distractions caused by young children. Except for the children variable, those variables that appear in more than one model have a consistent effect on the preference to telecommute.

The relatively small number of significant variables for each model (especially for the center vs. home preference model) may be due to the fact that the measures of preference for the two forms of telecommuting are highly correlated, which may have washed out the effects of some variables. Table 5-6 categorizes the respondents according to the center preference and home preference dependent variables. Nearly half of the sample prefers both the center and home forms of telecommuting. Of those who do not prefer the center, all but two prefer home, and of those who do not prefer home, all but two prefer the center. Thus, if a variable is related to center preference only, the high number of home preferrers who also prefer the center will blunt its effect. Ideally, one would construct a model to see which variables affect the preference of center only versus home only; however, as seen from Table 5-6, the sample size would be too small for this data set (15 and 27 respondents strictly prefer home and the center, respectively).

The center vs. home dependent variable is an attempt to force the respondents to choose a preference and still retain a majority of the data. Unfortunately (as shown in Table 5-7), half of the center preferrers for the third dependent variable are home preferrers according to the dependent variable for the first model, and half of the home preferrers are center preferrers according to the dependent variable for the second model. As a result, it is difficult to determine which variables are good separators of center and home preference. Nevertheless, the fact that the majority of the sample does not have an exclusive preference for one form of telecommuting over the other is a noteworthy empirical result. Realistic forecasts of telecommuting adoption should take this dual preference into account.

Return to the Abstracts of Non-Refereed Publications page.
Return to the Main page.