6. TRAVEL CHARACTERISTICS

6.1 Motivation and Methodology

In recent years, telecommuting has generated considerable interest in the research and planning community for its potential as an effective transportation demand management strategy (Mokhtarian, 1991). As the adoption of telecommuting becomes widespread, opportunities arise to evaluate telecommuting for its ability to alleviate congestion and improve air quality (Koenig, et al., 1996; Henderson, et al., 1996). The travel and emissions impacts related to travel associated with telecommuting centers are of particular interest and have been little-studied to date.

Center-based and home-based telecommuting are likely to differ considerably in terms of the resulting transportation and air quality impacts. An obvious difference concerns the reduction in the regular commute. While telecommuting from a center may reduce the length of the regular commute, home-based telecommuting may eliminate it altogether. However, there may be an increase in the number of discretionary trips made on home-based telecommuting days since there is more time for discretionary activities. Also, it is possible that a home-based telecommuter engages in several short home-other-home trips involving minimal trip chaining. Thus, we are likely to see some interesting tradeoffs between distance and the number of trips, both of which are important factors in modeling air quality impacts.

This chapter examines the travel characteristics of the sample of telecenter users investigated in this interim report. Four fundamental travel indicators ­ number of trips, person-miles traveled (PMT), vehicle-miles traveled (VMT, here taken to be miles traveled by driving alone in a personal vehicle), and commute mode choice ­ were studied. These indicators offer an overview of key travel characteristics and the analysis of vehicle-trips and -miles in particular lays the groundwork for conducting a future emissions and air quality analysis on these data. The ideal analysis would be a three-way study involving wave (before and after), group (telecommuters and non-telecommuters) and day type (telecommuting and non-telecommuting) as illustrated in Figure 6-1. Unfortunately, the restricted sample size precludes the possibility of such an analysis. A before-and-after study on the interim data set would have greatly reduced the sample size as only respondents common to both the before and after waves could be considered. Instead, we compared the four measures listed above between telecommuting (TC) and non-telecommuting (NTC) days in the after wave for telecenter users, with the non-telecommuting respondents treated as a control group. The main advantage offered by this methodology is that of using a larger sample size. The disadvantage, however, is that observed differences between telecommuting and non-telecommuting may be confounded by spill-over effects from one group into another (for example, in the form of deferred trips). These interactive effects would be controlled for in a before-and-after study.

The data used in the following analyses come from three sources: the after travel diary, the after attitudinal survey, and the sign-in log (all of which are described in Section 1.2). The travel diary provided data for the analysis of the first three parameters: number of trips, PMT, and VMT. On the other hand, the calculation of the commute mode choice distributions was relatively complex and involved input from multiple survey elements. The procedures involved will be discussed in detail in the appropriate sections. Before the analysis was carried out, the data set was cleaned and missing values were imputed where possible. On the after travel diary data, for example, the proportion of missing data ranged from none on most questions to nearly 8% for odometer readings. The proportion of missing data that could be corrected for the various questions ranged from none on the question regarding number of occupants to nearly 30% for the question on trip start times. These missing values were imputed by cross checking against other responses by the same respondent in the travel diary (for example, missing start times could be approximated from end times and an assumed average speed). Trip sequences were checked, and missing links were filled in to the extent possible. A total of 7 missing trips were added in order to complete trip records, one duplicated trip was deleted, and 18 trips with data entry errors were re-keyed.

6.2 Number of Trips, PMT, and VMT

In the following sections, the methodology used to study the first three travel indicators ­ number of trips, PMT, and VMT ­ is discussed, and the results of the analyses are presented. In Section 6.3, the methodology and the results associated with the study of mode choice distributions are presented.

6.2.1 Description of Travel Diary Data

The after travel diary data consists of 42 respondents providing information for 127 person-days (one person had four entries, all others were three-day diaries) and 586 total trips (includes trips made on not working days). The respondents fall into one of the three study groups depending on their telecommuting status: telecenter users, home-based telecommuters, and non-telecommuters.

The travel diary contained the following question for each day: "Today I am working from (check all that apply): primary office, telecenter, home, other location, not working" (see Appendix H). Based on the response to that question, each person-day was classified as one of four possible types as shown in Figure 6-2. A telecenter day (TC) was defined as a day on which a person worked from a telecenter irrespective of any other accompanying work location; this is illustrated in the figure by having any point in the dark gray circle classified as TC, even if the point also lies in the intersection with either of the other two circles. A day was classified as home-based telecommuting (HB) when the person worked only from home; thus in the figure, only points lying in the exclusive portion of the circle are classified as HB. A day was considered to be non-telecommuting (NTC) if a person worked from a regular work location and did not use the telecenter, even if s/he worked from home as well. And finally, a day was classified as a not working day (NW) if on that day the person did not work at all.

The elements of Figure 6-2 have been shaded to represent a hierarchy of expected direct travel benefits. Days on which no trips are made obviously represent the ideal travel reduction and are colored white. Next in order are NW and HB days. Both types of person-days are expected to provide similar travel improvements as the commute trip is eliminated. Between these two types, however, HB days are expected to result in greater travel reductions than NW days since presumably on an HB day a large proportion of the participant's waking hours are spent working. The next best person-day type would be the TC day, as the commute trip is considerably shortened. And finally, representing the highest-travel scenario is the regular NTC day, shaded black. It should be noted however, that the proposed hierarchy does not take into account indirect travel effects, such as deferred trips or a compensating effect leading to an increase in discretionary trips, that may alter the hypothetical ordering.

Some features of this classification are worth noting: there are days when persons worked from more than one work location, but the maximum number of such locations is two (on no day in this data set did a person work from all three work locations). Person-days classified as TC include not only days on which telecenter was the only work location but also days on which either home or the regular workplace was involved. This was done because in assessing the travel impacts of telecenters, it is important to take into account the extent to which telecenter use is accompanied by working from other locations. Classifying TC-HB days as TC is reasonable because the person has to make a commute trip to the center irrespective of the amount of time spent there. By classifying TC-NTC days as TC, we are conservatively estimating the actual effect of telecommuting centers on travel, but thereby guarding against the careless assumption that any telecenter occasion will automatically replace the normal commute. On the other hand, we classify as NTC days, those days on which both home and the regular workplace were work locations because a person makes a commute trip to the regular workplace irrespective of the amount of time spent there.

One alternative to the scheme adopted would be to define separate categories for each combination of work locations: TC, TC-NTC, TC-HB, HB, HB-NTC, NTC. This would however, make the comparison process rather cumbersome and considerably reduce the number of cases available in each group. In Figure 6-3, the row and column titles ­ RW, TC, and HM ­ on the work location matrices refer to the regular workplace, the telecenter, and the home respectively. The group of non-telecommuters do not have a work location matrix as they have only one work location, namely, the regular workplace. As is evident from the figure, the alternative classification scheme would lead to very small classes. Referring to the work location matrix for telecenter users, for example, there is only one person-day on which both a center and a regular workplace was used and only one on which the respondent worked from home and used the center.

From a travel perspective, a day on which a person does not work at all is quite different from a work day (whether working from a regular workplace, telecenter, or home.) There is no commute trip and if the person is not working because he or she is indisposed there may be no discretionary trips either. On the other hand, the person might have taken the day off specifically to engage in discretionary activities. To illustrate these differences, Table 6-1a compares the number of trips, PMT, and VMT measures for the three day types under consideration.

As can be seen from Table 6-1a (below), NW person-days are different from NTC days in terms of the average number of trips, PMT, and VMT. However, it could be anticipated that for this particular data set, pooling NW and NTC days would not affect the values of the travel parameters significantly. In fact, this has been the convention in previous studies. Table 6-1b presents the values for such a pooled group. By comparing the values from Tables 6-1a and b, we observe that the combined group has travel indicators that are slightly closer to those of the TC group than those of the NTC group alone. Thus, since combining the two groups provides more conservative estimates of the travel impacts of telecommuting, and for consistency with previous studies, we consolidate the NTC and NW groups in the analysis below. Should future studies involve a significant number of NW days, however, this day type should be analyzed separately in view of its distinctive characteristics.

¹ Mean calculated on a person-day basis.
² In the calculation of PV trips and VMT (vehicle miles traveled) only those trips which were made using a personal vehicle were considered.

For the remainder of this chapter, the NTC day type will include not working days. A final cross tabulation of the data set is given below in Table 6-2. As has been noted before, only the shaded cells, namely NTC and TC respondents on NTC and TC days, will be analyzed for the purposes of this report.

In the section that follows, we first study the differences between the control and study groups for the three travel indicators ­ number of trips, PMT, and VMT. Following that, in Section 6.2.3 we compare telecommuting and non-telecommuting days for telecenter users. In Section 6.2.4 we analyze the distributions of trips across time of day and purpose. In the following section we examine commute and non-commute trips separately. Finally, in Section 6.2.6 we estimate the average workday travel impacts for telecenter users by taking into account the frequency of telecommuting.

6.2.2 Control Group vs. Telecenter Users

If the impact of telecommuting on travel is to be correctly understood, the effects of confounding factors need to be accounted for. In other words we need to eliminate or control for characteristics which are extraneous to the process of telecommuting but which may affect the final outcome. To do this, the travel characteristics of telecenter users on NTC days were compared with the control group of non-telecommuters. Statistical tests were conducted to study if the differences in the measures were statistically significant. The mean and standard deviation of each of the three travel indicators for the two groups under study, namely non-telecommuters (control group) and telecommuters on non-telecommuting days, are given below in Table 6-3.

The average number of trips per person-day made by each group was found to be significantly different (p = 0.052), with the non-TC group making significantly more trips than the center-based telecommuters. On the other hand, the differences were insignificant (p = 0.265) if only personal-vehicle trips were considered, as is shown in the lower half of the table. The PMT values were also significantly different (p @ 0.000), with the TC group averaging a distance nearly 90% longer than the control group. The VMT figures show that the difference in average VMTs per person-day are also significant (p = 0.009) for the two groups. The average VMT per personal vehicle-day for the telecenter users is nearly 85% more than that for the control group.

As is evident from the tests conducted above, significant differences exist between the telecenter users and the controls, and the two groups are not exactly comparable. It is interesting to note that the relationships observed between the two groups in this study (fewer trips but higher miles traveled for telecenter users) mirror previous results reported by a study of data from the Puget Sound region (Henderson and Mokhtarian, 1996). That study hypothesized that the higher PMT (and VMT) for telecenter users is probably due to a longer average commute length, and the smaller number of trips is due to the longer commute trip taking up time that could otherwise be used for discretionary trips. Those hypotheses fit the current study as well. The average one-way commute distance to the regular workplace, obtained from the 39 telecenter users for whom after attitudinal survey data was available, was 44 miles (see Section 3.2.2), more than twice as long as the control group average of 18 miles (obtained from 10 respondents for whom data was available on both the after travel diary and the after attitudinal surveys).

6.2.3 Telecenter Users on TC vs. NTC Days

Since the non-telecommuter group is not as comparable to the telecenter users as would be desired, we focus on the comparison between telecommuting and non-telecommuting days for telecenter users. The comparisons for the number of trips, PMT, and VMT are shown in Table 6-4. As was done in the previous comparisons, t-tests were conducted to identify significant differences between the two sets of days.

Table 6-4: Comparison of the Number of Trips, PMT, and VMT for Telecenter Users on
Non-telecommuting and Telecommuting Days

The average number of trips per person-day was found to be statistically equal for the two day types (p = 0.819), at slightly more than four trips per day. The average number of trips made by personal vehicles, however, were significantly different (p = 0.004). The mean PMT per person-day was found to be significantly different for the two groups (p @ 0.000). The average PMT on TC days is more than 74% lower than on NTC days. As expected, the VMT indicator follows the outcome seen for PMT demonstrating significant differences between the two sets of days (p @ 0.000). A reduction of nearly 65% in average VMT per person-day on TC days was observed. Comparing the PMT to the VMT values and the values for average number of trips and average number of PV trips, the insignificant difference in the values for the TC day type is interesting. This seems to imply a very skewed mode choice distribution on telecommuting days, indicating that most telecenter users tend to employ their personal vehicles to make a majority of their trips on telecommuting days. This observation will be corroborated by findings in the mode choice sections (see Section 6.3).

From Section 6.2.2, the comparison between telecenter users and control group members shows that telecenter users have a larger average daily PMT (and VMT) on NTC days compared to the control group. In this section, the comparison reveals that these PMT and VMT figures fall considerably (74 and 65%, respectively) on telecommuting days. It is of interest to compare the travel activity of the telecenter users on telecommuting days to that of the non-telecommuters (see Table 6-5).

Table 6-5: Comparison of the Number of Trips, PMT, and VMT for Non-telecommuters and
Telecenter Users on Telecommuting Days

The difference between the average number of trips for the two groups is significant (p = 0.036), with the non-telecommuter group making a higher number of trips per person-day. The difference between the number of drive alone trips made by personal vehicles for the two groups is also significant in the same direction (p = 0.001). The PMT values are also significantly different (p = 0.003), with the telecenter users on TC days traveling nearly 52% shorter distances using all modes than non-TCers. The VMT values, however, are not significantly different between the two groups (p = 0.156), although VMT for telecenter users on TC days is about 35% lower, than non-TCers' VMT.

To summarize these comparisons, while telecommuters on ordinary commuting days travel, on average, 47% more miles than the control group members, they travel 52% less than the controls on TC days. Given that the non-telecommuters made more trips than telecommuters on NTC days, it is not surprising that they still made more trips than telecommuters on TC days. All three indicators, but especially PMT and VMT, point to considerable savings in travel, not just against telecommuters' own extreme baseline, but against a more typical employee's travel behavior on an average workday.

6.2.4 Trip Distribution Characteristics

Having studied the average number of trips per person-day by group and by day type, we now look at the number of trips in more detail by analyzing the impact of telecommuting on their distribution with respect to time of day and trip purpose. Next, the impact of telecommuting on trip chaining will be explored by comparing the number of links per trip chain.

6.2.4.1 Trip Distribution by Time of Day

The primary hypothesis with regard to the temporal distribution of trips is that telecommuters will reorganize their trips in order to avoid rush hour traffic. Figures 6-4a and b show the distribution of trips in our sample by time of day. The distribution has several interesting features that merit discussion. Considering first the trip start times, we see a clear ordering of the three groups, with the telecenter users on non-telecommuting days (TC-NTC) tending to start earliest (in the 5 - 7 AM window), the non-telecommuters (Non-TC) tending to start next and telecenter users on telecommuting days (TC-TC) tending to start latest (in the 7 - 8 AM window). This ordering is quite logical, with the members of the TC-NTC group starting first because of their long commute to work and the members of the TC-TC group starting last because of their short commute to the telecenter (see Section 3.2.2). The other interesting feature of the distribution is the significant lunch time peak for telecenter users on telecommuting days. This could be correlated with the significant proportion of eat meal and return home trips observed for this group (Purposes 2 and 6, see Figure 6-4c). Finally, the TC-TC group members seem to start their final return home trip in the evening before the other two groups, with very few of the telecenter users starting trips after 7 PM. In total it appears that travel on working days for the TC-TC group is considerably compressed (5 AM to 7 PM) in comparison with the TC-NTC group (4 AM to 9 PM) or the control group (5 AM to 9 PM).

Chi-squared tests of independence were performed to see if the distributions were statistically different between the three groups. A very small c² value indicates that the distributions compared are not significantly different, while a high c² value (and hence a small p-value) would indicate otherwise. To account for low cell counts in certain time windows, the categories were aggregated to two hour time windows with the tails of the distribution being aggregated into larger windows due to the sparse distribution of trips in the early morning and late night hours of the day. The aggregated distribution is given below in Table 6-6 and the test results are discussed below.

Table 6-6: Aggregated Trip Start and End Times

¹ This cell was originally empty but was set equal to 1 in order to calculate the c² statistic.

For trip start times, the distributions for the non-telecommuters and the telecenter users on regular commute days were found to be statistically identical (p = 0.068), while the distribution of trips on telecommuting and non-telecommuting days for telecenter users were found to be significantly different (p @ 0.000). Similar results were obtained for the trip end time distributions, with statistically identical distributions for the non-TC and TC-NTC groups (p = 0.398) and significantly different distributions for the TC-NTC and TC-TC groups (p = 0.029.) Thus, while non-telecommuters and telecenter users on non-telecommuting days tend to distribute their trips similarly, on telecommuting days the distribution is significantly different.

A previous study of home-based telecommuting (Pendyala, et al., 1991) found that the apparent shifts in the temporal distribution were actually due to overall reductions in trips occurring disproportionately in certain time periods. Specifically, commute trips were eliminated from the AM and PM peaks, thereby altering the relative distribution of trips, but trip-making for other purposes and in other time periods did not change significantly. PMT and VMT, on the other hand, were reduced across all time periods (Koenig, et al., 1996). Here by contrast, since the number of trips is not appreciably different between TC and NTC days, differences in the temporal distributions reflect actual shifting of trips. The implication is that telecenter users do move their trip start times around (by compressing their work day) in order to avoid the rush hour traffic, confirming the hypothesis stated earlier.

6.2.4.2 Trip Distribution by Purpose

Having explored the temporal distribution of trips, we will now discuss the distribution of trips according to purpose. Following a methodology similar to the one previously described, frequency distributions of trips according to purpose were obtained (see Figure 6-4c). Once again, several categories had to be aggregated due to low cell counts. The following categories were merged: 1 and 13; 3 and 14; and 9, 10, and 12. The aggregated frequency distribution and the results of the c² tests of independence are given below (see Table 6-7).

The following key will aid in interpreting Figure 6-4c:

     Purpose 1	Commute to Work			Purpose 8	Social / Recreation

     Purpose 2	Return Home			Purpose 9	Personal Business

     Purpose 3	Return to Work			Purpose 10	School / Education

     Purpose 4 	Work-related			Purpose 11	Change Mode

     Purpose 5	Drop Off / Pick Up Passenger	Purpose 12	Other

     Purpose 6	Eat Meal			Purpose 13	Commute to Telecenter

     Purpose 7	Shopping			Purpose 14	Return to Telecenter

Table 6-7: Aggregated Frequency Distribution of Trips by Purpose

¹ This cell was originally empty but was set equal to 1 in order to calculate the c² statistic.

The c² test of independence indicates that the distributions for the control and TC-NTC groups are significantly different (p = 0.039), and the TC-NTC and TC-TC groups are even more significantly different (p @ 0.000). Figure 6-4c brings to light some interesting details about the distributions. We see that there are significantly higher proportions of return home (2) and eat meal (6) trips on telecommuting days, as was discussed earlier. Also, there are no change mode trips on telecommuting days, while a significant proportion of the trips for the other two groups belong in the change mode purpose. The implication is that a smaller variety of modes are used on TC days, with most of the trips made using personal vehicles. This corroborates the conclusions drawn in Section 6.2.3 while comparing the travel indicators for non-telecommuters and telecenter users on telecommuting days. On telecommuting days we also observe higher proportions of shopping and social / recreation trips, which might have some interesting behavioral implications.

6.2.4.3 Trip Chaining

Finally, the extent of home-based trip chaining for telecenter users on telecommuting and non-telecommuting days was studied. The number of links in the home-to-home chain was used as an indicator of the degree of trip chaining. On non-telecommuting days there were 3.5 links per chain on average, compared to 2.7 links per chain on telecommuting days. This is consistent with the finding of Pendyala, et al. (1991) that the proportion of single-stop chains increased on telecommuting days. Perhaps, since the average commute on non-telecommuting days is rather long (79 miles), very few people tend to come home in the middle of the work day, and a considerable amount of trip chaining takes place on the long commute. On the other hand, on telecommuting days, the respondents tend to make a few more home-home chains with each chain having relatively fewer links since the telecenter is closer to home. Dividing by the average number of trips per person-day shown in Table 6-4 yields an average of 1.2 and 1.5 chains per day on NTC and TC days, respectively. A c² test shows that the distribution of the number of links (see Table 6-8) is significantly different on NTC and TC days (p = 0.049).

6.2.5 Comparison of Commute and Non-commute Travel

A detailed analysis of the after travel diary was performed to study the impact of telecommuting on both commute and non-commute travel. Since there is a potential for an increase in non-commute travel due to telecommuting (Salomon, 1985), the primary motivation for the analysis was to determine how the reduction in PMT and VMT, and the marginal reduction in trips (see Section 6.2.3), were distributed between commute and non-commute purposes.

A C program was developed to split PMT, VMT, and the number of trips per person-day into commute and non-commute purposes. To calculate commute PMT, the travel diary data were first scanned to check for direct home-to-work trips. If a direct home-to-work (or home-to-telecenter on telecommuting days) trip entry was present on any of the days, the corresponding distance was taken as the commute distance for that person. Otherwise, the commute distance reported in the attitudinal survey was used. Calculating the commute VMT was more complicated since it could vary by day for the same person, so it had to be identified separately for each trip. For a home-to-work sequence in which not all links were drive alone, the commute VMT was the minimum of the length of drive alone link(s) and the direct home-to-work commute distance from the attitudinal survey or from other travel diary days. The program calculates on an individual basis the total PMT, total VMT, total trips, and the number of commute trips per person-day (for a home-to-work-to-home chain, two commute trips are counted), using the following equations:

        non-commute trips/person-day = total trips/person-day - commute trips/person-day,

        non-commute PMT/person-day = total PMT/person-day - commute PMT/person-day,

        non-commute VMT/person-day = total VMT/person-day - commute VMT/person-day,

        commute PMT/person-day = commute trips/person-day × commute distance, and

        commute VMT/person-day = total commute VMT / total number of person-days,

The above measures were calculated for non-telecommuting days and telecommuting days (see Table 6-9). There is a drastic reduction in the commute PMT and VMT on telecommuting days which is not surprising since the commute distance of telecenter users to the regular workplace is much greater than to the telecenter. More interestingly, the table shows that the non-commute PMT actually decreases by almost a mile on telecommuting days, though the difference is not statistically significant (t = 0.14; p = 0.89). This is a positive result from a transportation viewpoint, which counters the hypothesis that non-commute travel increases on telecommuting days. However, though non-commute PMT decreases on telecommuting days, non-commute VMT actually increases by three miles on telecommuting days. Again, the difference is not statistically significant (p = 0.57). Figure 6-5 shows how the non-commute distance is distributed between drive-alone and all other modes on telecommuting and non-telecommuting days. There seems to be a decrease in the non-vehicular non-commute travel on telecommuting days.

Though the average numbers of trips on both non-telecommuting and telecommuting days are almost equal (t = 0.23; p = 0.82), the distribution of trips between commute and non-commute purposes is different. Contrary to original expectations, on telecommuting days, there is a statistically significant increase of 0.5 commute trips (t = 2.75; p = 0.01). The primary reason for this increase appears to be telecenter users going home for lunch more often on telecommuting days (see Section 6.2.4.2). (Going to the regular workplace on a telecommuting day was not a major effect, since that only occurred once in the sample as shown in Figure 6-3). The table also shows that there is a decrease of 0.6 non-commute trips on telecommuting days, though the difference is not statistically significant (t = 1.04; p = 0.29).

Table 6-9: The Impact of Telecommuting on Commute and
Non-commute Trips, PMT, and VMT¹

¹ Bolded means are significantly different between telecommuting and non-telecommuting days at a .LE. 0.05.

6.2.6 Aggregate Analysis

So far we have been analyzing the different day types separately. In the discussion that follows, various results are combined to obtain a more holistic view of the overall travel impacts of telecommuting for telecenter users. Telecommuting as a work option is not likely to replace conventional work schedules completely but will only occur for a certain percentage of days in a work week. To account for this, we compute the weighted average of travel indicators on TC and NTC days, where the weights are the relative frequencies of each type of day. Specifically:

where

GAGG is the aggregate value of a generic travel indicator (number of trips, PMT, etc.),
G ⁱ is the weighted average of the generic travel indicator for individual i,
N is the total number of respondents,
G ⁱXX is the simple average of the travel indicator for the day type XX (TC, NTC, or HB) for respondent i,
g ⁱXX is the value of the travel indicator for the ith respondent on some day of type XX,
N ⁱXX is the number of days of type XX for respondent i, and
f ⁱXX is the frequency of occurrence of day type XX for respondent i.

Figure 6-5: Distribution of Non-commute Distance Between Drive Alone and All Other Modes

Note that the value of GAGG as obtained from equation 6.1 above would not be the same as the value obtained by simply replacing each element of equation 6.2 with its sample average because, by averaging first at the individual level, one is accounting for any non-linear interactions that are likely to exist between the frequencies and the values of the travel indicators. For this study, however, the last term corresponding to telecommuting from home will be neglected as it constitutes a relatively insignificant portion (3 out of 72 person-days) of the telecenter users' person-days. Using equation 6.1, the aggregate values of the number of trips, PMT, and VMT are calculated (see Table 6-10). For non-telecommuters, f ⁱTC = f ⁱHB = 0, and G ⁱ = G ⁱNTC. For telecommuters, f ⁱHB @ 0 and f ⁱNTC = 1 - f ⁱTC. The values of the individual telecommuting frequencies used in these calculations have been taken from the six-month sign-in log average for each person (see Section 4.4).

Table 6-10: Comparison of the Aggregate Number of Trips, PMT, and VMT
for Non-telecommuters and Telecenter Users

The figures above indicate that at current telecommuting frequencies the aggregate average PMT and VMT are still significantly higher for center-based telecommuters than for non-telecommuters. This is because (1) the average non-telecommuting day PMT (and VMT) for telecenter users is considerably higher than for non-telecommuters (see Table 6-3), (2) the telecommuting frequency, f_TC, is not high enough to counter this difference in PMT (and VMT), and (3) neglecting home-based telecommuting inflates (albeit marginally) the proportion of non-telecommuting days.

The figures in Table 6-10 may be misleading in the sense that they seem to suggest that, in the aggregate, there are no positive travel impacts of telecommuting. However, in comparing the aggregate impacts of telecenter users to their own non-telecommuting baseline, the benefits of telecommuting for this group of long-distance commuters becomes clear. These results are shown in the final row of Table 6-10 for the no-telecommuting scenario. With the current levels of telecommuting, there is a reduction of more than 19% in average PMT (from 90 to 73 miles per person-day) when compared to the no-telecommuting alternative. Similar significant reductions can be observed in VMT (17%) too.

If no interactions existed between the frequency of telecommuting and the travel indicators, one could model the problem by taking a weighted average at the group level instead of the individual level. That is, the aggregate value of the travel indicator could be calculated as:

where

GXX is the value of the travel indicator averaged over all days of type XX and
FXX is the sample average frequency of the occurrence of day type XX.

Comparing the formulation of equation 6.1 with equation 6.3, it is apparent that the latter is biased since when there are interactions between travel and frequency, the mean of the product (that is, equation 6.1, the correct formulation) will not equal the product of the means. This approximate formulation, however, allows us to examine some interesting "what-if" scenarios. In the hypothetical results presented below, an FTC value of 0.182 (that is, an 18.2% frequency of telecommuting, obtained from the six-month sign-in log average), an FNTC value of 1 - 0.182 = 0.818, and an FHB value set equal to zero have been used. Due to the approximation alluded to above, the true values for the bolded statistics below are likely to be smaller for the first two and larger for the last one. Hence, they probably represent conservative bounds on the true values.

• The telecommuting frequency needed for the telecommuters to have a PMT equivalent to the non-telecommuters given the current travel parameters: 63.6%. That is, telecenter users in the sample would have to telecommute more than three days per week on average, in order for their aggregate PMT to decrease to the same level as that of the non-telecommuters in the sample.
• The non-telecommuting day PMT for telecenter users so that their aggregate PMT at the current average telecommuting frequency of 18.2% is equivalent to the non-telecommuters': 53.4 miles. That is, if telecenter users maintained their current average frequency of 18.2%, their non-telecommuting day PMT would need to average 53.4 miles (compared to 90.9 miles currently) in order for their aggregate PMT to drop to the level of the non-telecommuters.
• The reduction achieved in aggregate PMT assuming the same average VMT for telecommuters and non-telecommuters on non-telecommuting days: 9.4%. That is, if telecenter users' PMT on non-telecommuting days were equivalent to that of non-telecommuters (47.9 miles) and their telecommuting day PMT remained at 23.2 miles, their aggregate PMT would be 43.4 miles, a 9.4% reduction over the non-telecommuting baseline of 47.9 miles.

These hypothetical scenarios highlight the need for caution in extrapolating the observed results to the population as a whole. First, the question of selection bias should be addressed. Participation in the study as a non-telecommuting control group member was completely voluntary, and while recruitment was not rigorously random, travel indicators for this group appear to be reasonably representative. Participation as a telecenter user was also voluntary, and hence the data seem to suggest that telecommuting is most attractive to those with long commutes. This inference is quite intuitive and has been reported in other studies (Mokhtarian, et al., 1995).

However, while the evaluation team made no such request, it may be the case that employers selected telecommuting participants in this demonstration project partly on the basis of their commute length. This possibility, plus the fact that at least three-quarters of the non-telecommuting group expressed a desire to telecommute, suggest that commute length is not the only motivation in a preference to telecommute although it may be a key factor in the early adoption of telecommuting.

The implications for predicting and marketing telecommuting are important. From a market research perspective, the results would indicate that the potential market for early adopters of telecommuting may be primarily those people who live beyond a certain threshold distance from their regular workplace. However, more typical commuters should not be permanently neglected as a potential market.

As for the impacts of telecommuting on travel, two scenarios can be envisaged ­ one at each end of the continuum ­ with the reality falling somewhere in between. At one extreme is the scenario that telecommuting continues to be adopted primarily by long-distance commuters only. Even with this restriction, telecommuting has significant travel benefits because respondents commute much shorter distances to telecommuting centers than to the regular workplace as illustrated in Table 6-9. But when spread over the entire population, the benefits will be attenuated. At the other extreme is the scenario that telecommuting is adopted across the spectrum of commute lengths, such that the average commute length of telecommuters is equivalent to that of non-telecommuters. There too, telecommuting will have significant travel benefits, but the per capita reduction will not be as large as those seen here, perhaps closer to the 9% reduction in overall PMT per telecommuter estimated above for the hypothetical scenario in which telecommuters and non-telecommuters had the same VMT on non-telecommuting days, rather than the 19% observed for this sample of long-distance commuters. In either case, then, the ultimate population impacts of telecommuting will be lower than those suggested by this sample, either because telecommuting will only be adopted by a smaller segment of the population than initially envisioned or because the per-capita impacts will diminish as adopters become more representative, or both.

Thus, while the travel reductions observed here are clearly beneficial to the participating individuals, the system-wide effects will depend on how broadly telecommuting is adopted and by whom. In addition, other factors have not been and could not be addressed in this study, which could potentially mitigate the travel benefits; these factors include latent demand for travel and long-run impacts on land use.

6.3 Commute Mode Choice

This section analyzes the impact of telecommuting on the commute mode choice of telecenter users by comparing mode choices on telecommuting and non-telecommuting days. Two types of impacts may be hypothesized (Mokhtarian, 1991). First, on telecommuting days the proportion of transit and rideshare commute trips may be lower than on non-telecommuting days. This is because the commute trip to the telecenter is shorter and perhaps less well-served by the established transit systems and rideshare programs that focus on serving major employment centers. The second hypothesis is that commute trips to the telecenter (again because they are shorter) are more likely to involve environmentally-beneficial modes such as walk and bike.

The commute mode patterns of telecenter users are available from two primary sources, namely, the travel diaries and the attitudinal surveys. (In addition, commute modes on telecommuting days are identified in the sign-in log data, but distances traveled by each mode are not recorded there). Specific commute mode patterns are obtained from the travel diary. However, the travel diary data constitute only a particular snapshot in time, which for this small sample may not be representative of the behavior of the respondents. On the other hand, the attitudinal survey data may be more representative since it elicits an average commute mode pattern ­ actually the two most common patterns, with the percent of time each pattern is used. But, the data from the attitudinal survey may not be completely accurate since it relies on recall and since some respondents may use more than two patterns. Further, the attitudinal survey data may be more subject to respondent bias than the actual behavior recorded in the diaries, as suggested in Section 6.3.3. We should, therefore, expect differences in the commute mode choices obtained from the two sources.

A commute trip could consist of multiple trip segments and could have more than one mode. Two methods were used to calculate the mode splits: the primary-mode method and the distance-based method. In the first method, the mode used for the longest portion of the commute trip is identified as the primary mode, and the percent of trips for which a given mode is primary is calculated. In the second method, a weighted average of all modes used in any commute trip is calculated, where the weights are the distances for which a given mode is used. Also, the mode split analysis for both methods focuses only on the home-to-work trip since the trip to work is less likely to be contaminated with side trips than the trip home.

6.3.1 Commute Mode Choice from the Travel Diary Data

A total of twenty-four telecenter users completed the three-day after travel diaries. Thus, the total number of person-days was 72, out of which 39 were classified as non-telecommuting, 30 were telecenter-based telecommuting, and three were home-based telecommuting days (see Section 6.2.1 for the definition of each day type). However, not all non-telecommuting person-days involved a home-based commute trip. On two person-days, respondents did not work, and on two other person-days, the commute trip did not originate at home. Thus, the non-telecommuting day results presented below are based on the 35 person-days involving a regular commute trip. These 35 days actually comprise 36 commute trips, as individuals could go home and return to work sometime later in the day. Similarly, the 30 telecommuting person-days involved 34 commute trips to the telecenter.

A trip sequence was classified as a commute sequence if it originated at home and ended at either the regular workplace or the telecenter on the same person-day. Identifying the commute sequence was a non-trivial exercise since the number of segments in a commute sequence varies. A computer program was developed using the C language to identify the commute trips and evaluate the commute mode splits for telecenter users on telecommuting and non-telecommuting days.

Table 6-11 shows that on telecommuting days there is a substantial increase in the proportion of drive-alone commute trips. Also, as hypothesized, on telecommuting days the proportions of transit and rideshare commute trips decline, and the proportion of walk commute trips increases marginally. From the data, we can also determine that the number of segments per commute trip on telecommuting days (1.06) is lower than on non-telecommuting days (1.78). This is not surprising since the commute distance to the telecenter is significantly shorter than the commute distance to the regular workplace so there are fewer opportunities for trip-chaining. Another interesting observation was that on each of four (13%) telecommuting person-days, two commute trips were made to the telecenter. These additional trips home and back to work during the day are probably the result of having a shorter commute distance on telecommuting days. This result is discussed further in Sections 6.2.3 and 6.2.5.

Table 6-11: Commute Mode Split on Telecommuting and Non-telecommuting Days

¹ The following mode options were given in the diary but never selected by the respondents: drove/rode in electric vehicle, bus, light rail/trolley, and bicycle.

² Mode categories differed slightly between the travel diary and the attitudinal survey. To make the diary categories consistent with those on the attitudinal survey, the mode category "drove conventional motor vehicle" was split into "drive alone" and "carpool/vanpool" based on the number of people in the vehicle. The mode was considered to be "carpool/vanpool" if the number of people in the vehicle was greater than one. Also, the mode category "rode in conventional motor vehicle" was merged with the "carpool/vanpool" category.

It is interesting to note that the commute mode choices on telecommuting days obtained from the travel diaries are quite different from those obtained from the sign-in log data (Section 4.3.3). The percentage of drive-alone commute trips on telecommuting days obtained from the travel diaries (97.1%) is substantially higher than the percentage obtained from the sign-in log data (77.1%). A couple of reasons could explain this difference. Firstly, the sign-in log data is inherently weighted by the frequency of telecommuting whereas the travel diaries are less representative of actual frequencies. So, frequent telecommuters, who have a higher representation in the sign-in log data, could be making fewer drive-alone commute trips. Secondly, the two groups of people, travel diary respondents and sign-in log respondents, are not identical.

6.3.2 Commute Mode Choice from the Attitudinal Survey Data

6.3.2.1 NTC Day vs. TC Day Comparison

Data obtained from the 39 telecenter users who completed the after employee surveys were used to analyze the commute mode choices on telecommuting and non-telecommuting days. Questions D8 and D9 of the after employee survey ask for up to two patterns that telecenter users most often use to get to their regular workplace and telecommuting center respectively (see Appendix E). These questions obtain information on the mode used for each segment of the commute trip, the approximate length in miles of each segment, and the percent of time respondents use each pattern in terms of the total number of days that they commute to their regular workplace (or the total number of days that they work from the telecommuting center). Since the frequency with which each person works at home, the telecenter, and the regular workplace varies, it is important in obtaining aggregate mode choices to weight the above percentages by the frequency with which each individual commutes to the regular workplace and to the telecenter, respectively. For example, in calculating aggregate mode choices on telecommuting days, the pattern(s) of a person who telecommutes (from a center) 40% of the time should receive twice as much weight as those for a person who telecommutes 20% of the time.

As was discussed at length in Section 4.4.3, the frequency with which each respondent currently telecommutes from a center (fTC) is obtained from three sources, namely, the attitudinal survey, six-month sign-in log data, and one-month sign-in log data. The frequency with which each respondent currently telecommutes from home (fHM) is obtained from the attitudinal survey question (D11b, see Appendix E), "How much do you currently telecommute from home?" The frequency with which a person works from a regular workplace (fRW) is therefore: (1 - fTC - fHM). Tables 6-12 and 6-13 show the resulting commute mode splits obtained after weighting the data with the telecommuting frequencies using three data source combinations: AS (uses the attitudinal survey data for modes and fTC), SIL6 (uses the attitudinal survey for modes and the six-month sign-in log data for fTC), and SIL1 (uses the attitudinal survey for modes and the one-month sign-in log data for fTC).

Table 6-12: Primary Mode Split by Data Source

From the above tables, it is clear that the percentage of drive-alone commute trips is substantially greater on telecommuting days than on non-telecommuting days. An interesting observation is thatthere is a slight increase in the proportion of carpool commute trips on telecommuting days. But, the overall rideshare (including vanpool) commute trips on telecommuting days has decreased. Also, there is a slight increase in the proportion of commute trips by walk on telecommuting days. A chi-squared test of the distributions shows no statistically significant differences among the mode splits obtained using the three data sets.

6.3.2.2 Before vs. After Comparison

A before and after comparison of the commute mode choices as reported in the employee surveys was done for the 27 respondents completing both surveys. In the case of non-telecommuting days, the comparison is a "true" before and after comparison of commute mode choices with a time interval of approximately six months. In other words, the analysis will reveal whether telecommuting has affected the commute mode choices of telecenter users on non-telecommuting days. It has been hypothesized, for example, that telecommuting might lead to the dissolution of carpools, resulting in more drive alone trips even on non-telecommuting days (Mokhtarian, 1991).

The before and after comparison on telecommuting days is actually a prospective vs. current comparison of commute mode choices. In the before employee surveys, the respondents were asked to give the commute mode pattern that they plan to use most often to get to the telecommuting center. The before commute mode pattern on telecommuting days was evaluated by weighting this pattern with the frequency of telecommuting obtained from the question, "Six months from now, how much do you expect to be telecommuting from a telecommuting center?" The after commute mode pattern was weighted by the current frequency of telecommuting reported in the attitudinal survey. Tables 6-14 and 6-15 show the comparison on both non-telecommuting days and telecommuting days.

From the following tables, the before and after commute mode splits are almost identical on non-telecommuting days, implying that telecommuting has not affected the commute mode choices of the respondents on non-telecommuting days. Also, the prospective and current commute mode choices on telecommuting days are very similar suggesting that the respondents had a fairly good idea (before they actually started telecommuting) of the commute modes they would be using to get to the telecenter.

6.3.3 Comparison of Travel Diary vs. Attitudinal Survey Mode Split

The commute mode splits obtained from the two sources, namely travel diaries and after employee surveys, are not very similar. The travel diaries show a substantially greater percentage of drive-alone commute trips on telecommuting days. Conversely, the surveys show some expected use of light rail and some expected and actual use of bus on non-telecommuting days, whereas the travel diaries do not record any uses of these two modes. Several reasons could explain these differences. First, the two samples are not identical: only 22 people fall in the intersection of the 24 diary respondents and the 39 survey respondents. Second, the bus and light rail modes were used (or expected to be used) so infrequently that they may legitimately not have been used during the three-day travel diary period. Finally, it is also possible that respondents are more likely (consciously or subconsciously) to overstate their use of environmentally correct modes for the general question on the attitudinal survey than to deliberately falsify their actual behavior as recorded on the diaries.

6.4 Summary of Travel Characteristics

In this chapter the travel characteristics of the respondents were studied. Four main travel indicators, namely the number of trips, PMT (person-miles traveled), VMT (vehicle-miles traveled), and mode choice distributions, were studied. Two main sets of comparisons were made: the first between the control group of non-telecommuters and telecenter users on non-telecommuting days and the other between telecommuting and non-telecommuting days for telecenter users.

The control group was found to have significantly different travel characteristics than the telecommuting group, with the latter making fewer trips (4.3 compared to 5.9) but traveling longer distances (90.9 average weekday PMT compared to 47.9 miles for the control group), on average. The differences in PMT and VMT could be attributed to differences in commute distances, and the difference in number of trips could be linked to the fact that telecenter users are left with a significantly smaller amount of time for discretionary activities in view of their long commutes to the regular workplace.

Since the non-telecommuter group is not as comparable to the telecenter users as would be desired, we focus on the comparison between telecommuting days and non-telecommuting days for telecenter users. Comparing telecommuting days and non-telecommuting days, one finds that while the average number of trips are almost the same, PMT and VMT values are significantly different, with the average weekday distance traveled by all modes decreasing by more than 74% on telecommuting days. Also, telecommuters on telecommuting days showed a reduction of nearly 52% in PMT when compared with the controls. Thus, these results point to considerable savings in travel on telecommuting days, not just against telecommuters' own extreme baseline, but against a more normal employee's travel behavior on an average workday.

Next, the distribution of trips with respect to time of day and purpose was explored. Significant differences were found between the temporal distributions of trips on telecommuting days and non-telecommuting days. Comparisons of telecommuters' non-telecommuting days with the control group, however, showed no significant differences. The distributions exhibited an interesting ordering of trip start times with telecenter users on non-telecommuting days starting the earliest, followed by the control group, and then telecenter users on telecommuting days (because of their significantly shorter commute distances). The temporal distribution for telecenter users on telecommuting days also showed a significant lunch time peak. Comparisons of the distribution of trip purposes indicated that the distributions for the control group and the telecenter users on non-telecommuting days were statistically different while those on telecommuting and non-telecommuting days were even more significantly different. The differences arose due to (1) a significantly higher number of return home and eat meal trips, (2) a higher proportion of shopping and social/recreation trips, and (3) an absence of change mode trips on telecommuting days. This last factor corroborates the conclusions drawn by comparing the travel indicators for non-telecommuters and telecenter users on telecommuting days (see Section 6.2.3) and the discussion on mode choice (Section 6.3), and implies that a smaller variety of modes are used on telecommuting days, with most of the trips made while driving alone.

To study the trip chaining behavior of the respondents, the average number of links in a home-home chain were compared for telecommuting and non-telecommuting days. The comparison revealed a significantly higher number of links on non-telecommuting days. This could be attributed to the long commute distances on such days. As the average number of trips is almost the same for the two sets of days, one could hypothesize that on telecommuting days the respondents make a larger number of home-to-home cycles involving a smaller number of links.

A comparison of commute and non-commute travel on telecommuting and non-telecommuting days showed that there is a drastic reduction in the commute PMT (by 66.9 miles) and VMT (by 41.0 miles) on telecommuting days. Also, the non-commute PMT decreases almost a mile on telecommuting days. However, non-commute VMT actually increases by more than two and a half miles on telecommuting days. Therefore, there is a decrease in the non-vehicular, non-commute travel occurring on telecommuting days.

Though the average numbers of trips on both non-telecommuting days and telecommuting days are almost equal, the distribution of trips between commute and non-commute purposes differs. On telecommuting days, there is a statistically significant increase of 0.5 commute trips. Also, there is a decrease of 0.6 non-commute trips on telecommuting days, though the difference is not statistically significant.

Next, the commute mode choice distributions for the study groups were analyzed. The travel diary and attitudinal survey data show that there is a substantial difference between the commute mode choices of telecenter users on telecommuting days and non-telecommuting days. The percentage of drive-alone trips is substantially higher and the percentages of transit and rideshare trips are substantially lower on telecommuting than on non-telecommuting days. Also, a before and after comparison of the commute mode splits reported in the attitudinal survey revealed that telecommuting has not affected the commute mode choices of the respondents on non-telecommuting days.

Finally, to obtain a better understanding of the overall process, the aggregate values of the travel indicators were studied. This was done by weighting the travel indicators by the corresponding telecommuting frequency. The aggregate figures indicate that at current frequencies of telecommuting (18.2% on average, or approximately once in five days), the telecenter users travel significantly larger distances: a composite weekday average of 73.1 miles compared to 49.1 miles for the control group. Two factors contribute to the difference in aggregate travel between telecenter users and the controls: (1) the non-telecommuting day PMT (VMT) for the telecenter users is considerably larger than that for the controls, and (2) the level of telecommuting is not high enough to counter this difference. But, while the telecenter users still travel more than the control group members, they would have had an average PMT of 90.9 miles had they not been telecommuting. Telecommuting from a center reduced their total weekday travel nearly 19%.

It could be hypothesized that commute distance is an important factor in the preference to telecommute and that a self-selection bias occurred in the selection of the study groups, with respondents who lived farther away opting to be in the telecommuting group. However, the possibility that commute distance might have been a criterion used by the employers in selecting respondents for the telecommuting group (thus generating an unintentional bias) and the fact that at least three-quarters of the non-telecommuting group expressed a desire to telecommute suggest that commute length is not the only motivation in a preference to telecommute. If telecommuting is primarily attractive to long-distance commuters, considerable per capita reductions in travel will result, though only in a particular market segment. Conversely, if the appeal is more universal, the reductions in travel per capita are not likely to be as high as in this sample but would apply to a larger segment of the workforce.

Previous
Next
Table of Contents

Return to the Abstracts of Non-Refereed Publications page.
Return to the Telecommunications and Travel Main page.
Return to the Teleworking Reports page.