4. TELECOMMUTING PATTERNS

4.1 Introduction

Though many studies have examined attitudes toward telecommuting, preference for telecommuting (Mokhtarian and Salomon, 1997; Stanek, 1995; Bagley, 1995), choice of telecommuting (Mokhtarian and Salomon, 1996; Bernardino, et al., 1993; Mahmassani, et al., 1993), and characteristics of telecommuters (Hartman, et al., 1991; Yap and Tng, 1990), few studies to date have attempted to explore patterns of telecommuting behavior in detail. Questions of interest include how often individuals telecommute, the duration of their telecommuting participation, how much time they spend at the telecenter, and how they distribute their time over various work locations on a given day.

It is useful to analyze these types of questions both at the disaggregate and aggregate levels. At the disaggregate level, analyzing telecommuting behavior provides further insight into individual choice patterns, and offers the potential for building models to explain and predict those choices. On the aggregate (site specific and all sites combined) level, such an analysis will support the evaluation of the centers' operational performance, the development of marketing strategies, and the recruitment of participants.

The following section discusses data collection and cleaning procedures. Section 4.3 presents the aggregate analysis of telecommuting patterns, including: utilization of the telecenter, work time spent at the telecenter, workplace combinations on telecommuting days, and mode of travel to the telecenter. Section 4.4 offers a disaggregate analysis of telecommuting patterns, including telecommuting duration, telecommuting frequency, proportion of telecenter-only working days, and mode choice to the telecenter. Section 4.5 summarizes the key findings of the chapter.

4.2 Data Collection and Cleaning

4.2.1 Data Collection

The information was collected at the telecenters participating in the project. Telecommuters were instructed to make an entry in the attendance log each day they used the telecenter. The entry included date, name, transportation mode used to get to the telecenter, and estimated work time to be spent at various workplaces, including telecenter, main office, home, and any other work location. An example attendance log sheet is found in Appendix G.

To the extent participants forgot or declined to sign in at each use, these data may somewhat undercount the usage of the telecenter by telecommuters. However, site administrators had an incentive to ensure the most accurate reporting possible, as occupancy levels were calculated based on the sign-in data and each site had a contractual obligation to meet certain occupancy levels (see the companion volume on telecenter operation). In addition, other uses of each telecenter occurred which were not captured by the attendance log (see Section 2.3).

For this interim report, only sign-in data through June 30, 1995 were analyzed (for Davis and Anaheim, no data were available after January and March 1995, respectively). An additional year of sign-in data (through June 30, 1996) will be available for analysis in the final report.

Table 4-1 lists the availability of attendance log data at the RABO and non-RABO telecenters that provided attendance data. A majority of the data comes from the two non-RABO sites since they have been operating for a longer period. The data availability according to each site is shown in Table 4-2. The information from the telecenter in Ontario dominates the overall data with 46.7% of all telecommuting occasions. More than half of the information gathered at RABO sites comes from the telecenters in Coronado, Grass Valley, Modesto, and Chula Vista. The amount of data produced by each site depends on (1) the operating length of the center, (2) the number of telecommuters, and (3) the frequency of use by each telecommuter.

Table 4-1: Data Availability

¹ Denotes total number of person-day telecommuting occasions.
² At the Ontario center, the number of workstations decreased by six on March 1, 1994.

4.2.2 Data Cleaning

Each telecommuter was expected to sign in once for each telecommuting occasion. However, even after correcting data entry errors, double entries in the data set occurred due to the following reasons: (1) the telecommuter signed in twice on the same day (this may be the result of the telecommuter returning to the telecenter later in the day and forgetting that s/he had signed in already) or (2) the telecommuter attended two different telecenters on the same day.

Several rules were defined to process these unusual cases:

• If the two entries made on the same day by an individual were identical, one of them was eliminated. In the original data set, 26 cases (0.4% of RABO and non-RABO cases combined) fell into this category.
• If the two attendance log entries were for the same telecenter and different from each other, the one with the longer work time spent at the telecenter was preserved. It is assumed that the longer duration was the sum of work time for the two visits. Three such cases were found in the original data set.
• For the occasions on which a telecommuter signed in at two different telecenters on the same day, both records were preserved for aggregate analyses of measures such as site occupancy rate. However, for the disaggregate analysis of indicators such as telecommuting frequency, the one with longer time spent at the telecenter was selected. Six of these cases were found.

Some entries in the data set are missing. The non-RABO sites of Highland and Ontario did not begin using the attendance logs designed by this project until February 1, 1994 and September 1, 1994, respectively. Prior to these dates, those sites obtained sign-in and sign-out times for each telecommuter but did not request information on transportation mode used to get to the telecenter nor on the distribution of work time at locations other than the telecenter. Thus, this information was not available for 2774 (66%) of the 4206 telecommuting occasions recorded in Table 4-1. Among these 4206 occasions, even the information on work time at the telecenter was missing for 1566 cases (37.2%).

Site	Start Date	End Date	Number of Weeks	Number of Occasions1	Number of Users	Number of Workstations
Coronado	11/01/93	06/30/95	86.6	278 (4.2%)	15	4
Grass Valley	02/08/94	06/30/95	72.4	515 (7.8%)	10	6
Anaheim	06/30/94	03/01/95	34.9	108 (1.6%)	10	15
Vacaville - Ulatis	08/01/94	06/30/95	47.6	230 (3.5%)	25	7
Davis	08/11/94	01/12/95	22.0	15 (0.2%)	4	10
Vacaville - Alamo	08/15/94	06/30/95	45.6	231 (3.5%)	20	8
Chula Vista - H St.	09/19/94	06/30/95	40.6	410 (6.2%)	15	10
Modesto	10/18/94	06/30/95	36.4	235 (3.6%)	10	10
Chula Vista - F St.	11/01/94	06/30/95	34.4	166 (2.5%)	7	8
Ventura Community College	02/01/95	06/30/95	21.3	87 (1.3%)	3	5
La Mesa	03/07/95	06/30/95	16.4	36 (0.5%)	5	6
Moorpark Community College	04/17/95	06/30/95	10.6	60 (0.9%)	2	5
Ontario	11/27/91	06/30/95	187.3	3071 (46.7%)	157	24 (182)
Highland	12/08/92	06/30/95	133.4	1135 (17.3%)	18	6
Total				5577 (100%)	2913	124 (1182)

¹ Denotes total number of person-day telecommuting occasions.
² Eighteen workstations were available after March 1, 1994.
³ Denotes total number of individuals: 9 telecommuters attended both of the Vacaville telecenters and 1 telecommuter attended both of the Chula Vista telecenters. They are counted under each site but not double-counted in the total.

Among the 2371 entries from RABO sites, the information associated with transportation mode and distribution of work time was missing for 65 (2.7%) and 23 (1.0%) of the cases, respectively. Among the 1432 entries from non-RABO sites using the attendance logs designed for this project, the information about transportation mode and distribution of work time was missing for 132 (9.2%) and 134 (9.4%) of the cases, respectively. The amount of missing data associated with the information for each site is shown in Table 4-3. Finally, attendance logs for some months are missing from some telecenters because the site administrators failed to provide the information: specifically August 1993 for Ontario, January 1994 for Highland, and September and October 1994 for Davis. No attempt was made to estimate the number of telecommuting occasions or any other information for these months.

¹ N is the total number of telecommuting occasions.

4.3 Aggregate Analysis

In this section, aggregate telecommuting patterns both across all sites combined and by each site separately are presented. The following four sections respectively discuss utilization of the telecenter, work time spent at the telecenter, workplace combinations on telecommuting days, and means of travel to the telecenter.

4.3.1 Utilization of the Telecenter

To evaluate the operational performance of telecenters, one must develop a method to express how the facilities were utilized. The total number of telecommuting occasions is not an adequate measure because the telecenters are of different sizes. Thus, the largest number of occasions is likely to occur at the telecenters equipped with the most work stations. To control for differing sizes, two measures of telecenter utilization, the usage rate and the occupancy rate, were developed.

The monthly telecenter usage rate is the total number of occasions (person-days) on which the center was used for telecommuting, divided by the product of the number of work stations and the number of working days in the month:

usage rate = (# of telecommuting occasions)/(# of workstations x # of work days) (4.1)

The monthly telecenter occupancy rate is calculated as the number of telecommuting occasions that are at least four hours long divided by the same denominator:

occupancy rate = (# of telecommuting occasions of at least 4 hours)/(# of workstations x number of work days) (4.2)

That is, the usage rate is the proportion of "workspace-days" for which the center was used for any length of time (for telecommuting), and the occupancy rate is the proportion of workspace-days for which it was occupied at least four hours. These formulas draw on the concepts of exposure as used in accident studies. The denominator in equation 4.1 could be interpreted as the total number of possible opportunities for telecommuters to be exposed to the telecenter.

The total number of working days is adopted instead of the total number of days in the month because we are focusing on telecommuting as a substitution for commuting to the office on a normal working day. For the purposes of this analysis, "working days" excludes Saturdays, Sundays and eight federal holidays (New Year's Day, Martin Luther King's Birthday, President's Day, Memorial Day, Independence Day, Labor Day, Thanksgiving Day, and Christmas). For the aggregate analysis presented here in Section 4.3, if the initial observation for a center did not coincide with the first day of the month, only working days from the first telecommuting occasion onward were included in the usage and occupancy rates calculated for that first month of operation. For the disaggregate analysis presented in Section 4.4, if the initial observation for an individual did not coincide with the first day of the month, only working days from the first telecommuting occasion onward were included in the corresponding calculation for that individual.

Figure 4-1 shows the average usage and occupancy rates across all RABO sites, from the opening of the first site (Coronado) in November 1993 through June 1995, the cutoff point for analysis for this interim report. The trend varies considerably over time. Several reasons make it difficult to explain or predict the variation. First, the telecenters opened at different times. Since a center will typically open with relatively low usage and then build up over time, entry of a new site into the calculation tends to depress the average. (The bottom line on Figure 4-1 indicates how many sites were open each month, that is, the number of sites over which the average rates were calculated). Further, the size of each center also affects usage and occupancy rates. Fifty telecommuting occasions in a week would represent 100% usage for a 10-workstation site, but only 50% usage for a site with 20 workstations. Finally, the rates also change according to the number of telecommuters at the site in that month, as well as the number of their telecommuting occasions. Some telecommuters quit telecommuting, some newly joined, and some did not telecommute during certain months.

In spite of these factors, a couple of tentative observations may be made. First, the rates appear to have stabilized somewhat for the last 6 months of observation: January 1995 to June 1995. The average usage rates ranged between 15 and 20%. The occupancy rates were somewhat lower ranging from 10 to 15%. Secondly, there does appear to be a seasonal effect, specifically a "summer slump". The dip in the summer of 1994 is partially confounded by the entry of four new sites, but the trend (based on only two sites) appears to have pre-dated those new openings. Summer 1995 data are not analyzed for this report, but Figure 4-1 seems to indicate a coming dip based on the June 1995 data point. It will be important to combine these data with the additional year of operation that will be available for the final report in order to solidify these conclusions.

To control for the confounding factor of sites opening at different times, average usage and occupancy rates were computed based on the number of months a site had been open. Note, however, that this introduces the confounding factor of seasonality. That is, the summer and winter holiday seasons will occur after a differing number of months of operation for each center, so it would not be easy to separate out those effects. The only way to control for both start-up and seasonal effects would be to extend the time period of observation well past the point at which start-up effects would be negligible for all sites and then look at the average rates on a calendar-month basis as in Figure 4-1.

Figure 4-2 shows the average usage and occupancy rates for the first 10 months of operation across all RABO sites. The average usage rate begins at 9% in the first month of operation and rises to 16% after 10 months; the average occupancy rate rises from 9 to 12% over the same period. The trends show slow but steady growth in the utilization of the telecenters. However, as shown in Figure 4-3, the individual usage rates demonstrate significant variation not only between sites but also at each site.

Figure 4-4 shows the usage rates at non-RABO sites. The center in Ontario maintained roughly steady usage rates between 10 to 20%. In contrast, Highland had a relatively high utilization, especially over the last few months of observation (from 40 to 62%). However, this relatively high usage was due to (1) the small number of workstations (six) compared to Ontario (24 or 18); and (2) the fact that several telecommuters, including a real estate agent, used the center nearly every day. Obviously, the usage rate is a function of the number of the workstations at the center and number of person-day telecommuting occasions.

4.3.2 Work Time Spent at the Telecenter

On each telecommuting occasion, telecommuters reported how long they worked at the telecenter. The average work time for RABO telecenter users was 5.73 hours, with a standard deviation of 2.72 hours. Non-RABO telecenter users stayed at the telecenter for 6.73 hours on average, with a standard deviation of 2.83 hours. Under a test for equality of means, the two average work times were statistically different (p @ 0.000).

Figure 4-5 illustrates the distribution of reported work time spent at the telecenter. For the RABO sites, peaks appear at 3-4 hours and 7-8 hours. These peaks reflect the tendency to spend either half a day or a full day at the telecenter. About 56% of the occasions fell into these two categories. Nearly 51% of the telecommuting occasions lasted 6 hours or longer. Although 37.2% of the information was missing for non-RABO sites, the available data show that the telecommuters were likely to work only at the telecenter on their telecommuting days. Figure 4-6 shows the cumulative distribution of the work time spent at the telecenter for both RABO and non-RABO sites based on the available data. A c² test shows that the two work time distributions are significantly different (p @ 0.000), with RABO sites showing a higher proportion of shorter telecommuting occasions.

4.3.3 Workplace Combinations on Telecommuting Days

On telecommuting days, it is possible that the telecommuters worked at more than one location, such as the regular workplace (RW), home (HM), and other locations (OL) as well as the telecenter (TC). To better understand how the telecenter is used, it is desirable to analyze the frequency of various workplace combinations. Eight combinations are possible: (1) TC only, (2) TC/RW, (3) TC/HM, (4)TC/OL, (5) TC/RW/HM, (6) TC/RW/OL, (7) TC/HM/OL, and (8) TC/RW/HM/OL.

Figure 4-7 shows the distribution of telecommuting occasions for each of the eight workplace combinations at RABO sites and non-RABO sites. At RABO sites, the most common patterns are (1) TC only (59.8%), (2) TC/OL (14.9%), and (3) TC/HM (9.3%). For the combinations involving the center and one or more other locations, 10.6% of all RABO telecommuting occasions involved working from the regular workplace, 16.8% involved working at home, and 23.9% involved working from another location. Overall, only 10.4% of RABO occasions involved working at more than two locations.

Working solely at the telecenter was by far the most frequent telecommuting pattern at non-RABO sites (85.1%). The second and third most common workplace combinations were TC/OL (8.7%) and TC/RW (4.1%), respectively. Contrary to expectation, telecommuting from a center is not often combined with home-based telecommuting in the same day. Although many participants state they wish to engage in both home-based and center-based telecommuting to some degree (see Section 3.2.4), apparently any particular day is more likely to see one or the other forms of telecommuting exclusively. However, telecenter/home combinations are common for the five users of the newer sites in Moorpark and Ventura. Therefore, the aggregate pattern may change with a longer operating history for those two sites.

The average work time distributed by workplace combination for RABO and non-RABO sites is presented in Tables 4-4 and 4-5, respectively. The findings for the most common combinations can be summarized as follows.

• On average, nearly 7 hours were spent at the center if the telecenter was the only work location used that day.
• If the telecommuters worked at the center and at an "other" location (TC/OL), the work time spent at the telecenter dropped to about 3.5 hours on average. More work time was spent at the other workplace than at the telecenter. It is likely that the telecommuters' main task on that day was to work at the other location. They may have gone to the telecenter to prepare for or finish that main task.

Table 4-6 shows that the distribution of telecommuting occasions for the eight workplace combinations varies considerably across sites. Except for Alamo, a plurality of the telecommuting occasions at each site was telecenter-only; this was the majority pattern at eight of the fourteen sites. Working at the telecenter and an other location (TC/OL) was popular at Ulatis, Alamo, and Chula Vista (H St.). This is in keeping with the use of the Ulatis and Alamo sites by several health care workers, who spent part of the day rendering services at patients' homes.

The TC/HM pattern was common at Coronado, Moorpark Community College, and Ventura Community College. The TC/RW and TC/HM/OL patterns were fairly popular at Anaheim and at Chula Vista (F St.), respectively. The variation in workplace combinations among sites is likely the result of the diverse job characteristics of telecommuters.

At non-RABO sites, the telecenter-only pattern dominates the sample, occurring on about 85% of the occasions. The TC/OL and TC/RW are the second most common patterns at Highland and at Ontario, respectively.

4.3.4 Travel Modes for Accessing the Telecenter

Table 4-7 depicts the distribution of the primary transportation mode used to access RABO and non-RABO telecenters. For more than three-fourths of the telecommuting occasions (77.1%) at RABO sites, telecommuters drove alone to the telecenter. Other common travel modes were being dropped off (7.3%), carpooling (6.0%), and walking/biking (5.6%). On 0.9% of the occasions, an alternative-fuel vehicle was used (most of these trips were made to the Moorpark site). Clearly, driving alone was the primary means of commuting to RABO sites.

The utilization of transportation modes at non-RABO sites is similar to that at RABO sites. Driving alone was even more common for non-RABO sites (89.2%) than for RABO sites. Thus, placing telecenters near residential areas may have had a marginal effect on lowering the share of drive-alone access trips. However, driving alone remained the dominant mode in both cases. For a more extensive discussion of the transportation impacts of telecenters, see Chapter 6.

Table 4-4: Work Time Spent at Each Workplace for Various Combinations (RABO Sites)

Workplace Combination	Number (Proportion)	Hours (Mean and Standard Deviation)
		Telecenter	Regular Workplace	Home	Other Location
TC	1402 (59.8%)	6.92 (2.41)	---	---	---
TC/RW	131 (5.6%)	4.44 (1.89)	3.84 (1.79)	---	---
TC/HM	218 (9.3%)	5.25 (2.08)	---	2.98 (1.62)	---
TC/OL	349 (14.9%)	3.53 (2.16)	---	---	4.73 (2.14)
TC/RW/HM	32 (1.4%)	3.34 (1.60)	3.20 (1.95)	2.05 (1.22)	---
TC/RW/OL	69 (2.9%)	2.58 (1.63)	2.61 (1.53)	---	3.48 (2.14)
TC/HM/OL	126 (5.4%)	3.55 (1.45)	---	1.66 (1.04)	2.82 (2.08)
ALL	17 (0.7%)	2.46 (1.57)	1.26 (0.53)	1.53 (0.78)	3.91 (2.45)
Total RABO	2344	5.73 (2.72)	0.34 (1.16)	0.41 (1.10)	0.99 (2.08)

Table 4-5: Work Time Spent at Each Workplace for Various Combinations (Non-RABO Sites and Overall)

¹ The proportion is based on data from the non-RABO sites only.

Site	N2	Workplace Combination3
		TC	TC/RW	TC/HM	TC/OL	TC/RW/HM	TC/RW/OL	TC/HM/OL	ALL
Coronado	274	63.9%	4.4%	19.7%	3.6%	3.6%	3.6%	1.1%	0
Grass Valley	515	85.4%	4.3%	5.6%	2.3%	0	1.0%	0.2%	0
Anaheim	108	68.2%	22.4%	0.9%	7.5%	0.9%	0	0	0
Vacaville - Ulatis	223	46.6%	2.2%	3.6%	37.2%	0	4.0%	2.2%	4.0%
Davis	15	86.7%	6.7%	6.7%	0	0	0	0	0
Vacaville - Alamo	228	32.5%	0	3.9%	39.0%	0.4%	7.0%	15.4%	1.8%
Chula Vista - H St.	405	36.0%	13.1%	11.1%	27.2%	1.7%	4.9%	5.7%	0.2%
Modesto	234	96.6%	0	2.6%	0	0	0	0.9%	0
Chula Vista - F St.	165	38.8%	1.2%	15.2%	9.7%	1.2%	1.2%	31.5%	1.2%
Ventura	87	39.1%	12.6%	24.1%	3.4%	5.7%	8.0%	5.7%	1.1%
La Mesa	36	82.9%	2.9%	0	14.3%	0	0	0	0
Moorpark	60	43.3%	0	31.7%	25.0%	0	0	0	0
Highland	734	83.7%	1.4%	1.0%	12.4%	0	0.3%	1.2%	0
Ontario	564	87.1%	7.4%	0.9%	3.9%	0.4%	0.4%	0	0
All Sites	36484	68.9%	5.0%	6.3%	12.7%	0.9%	2.0%	3.7%	0.5%

¹ The most frequent combination for each site is bolded, and the second most frequent is italicized and underlined.
² N is the number of telecommuting occasions.
³ Workplace locations are the regular workplace (RW), the telecommuting center (TC), home (HM), and other location (OL).
⁴ This number differs from the total of Table 4-5 in that 6 occasions involved both Vacaville sites and are here counted under both sites.

Site	N	Walk/ Bike	Alt.-Fuel Vehicle	Drive Alone	Carpool	Public Transit	Dropped Off	Other Mode
Coronado	272	24.6%	1.1%	57.7%	12.9%	0.4%	2.6%	0.7%
Grass Valley	501	4.4%	0	72.1%	0.2%	0.2%	23.2%	0
Anaheim	105	0	0	77.1%	21.9%	0	1.0%	0
Vacaville - Ulatis	226	0.9%	0	95.6%	3.5%	0	0	0
Davis	15	26.7%	0	73.3%	0	0	0	0
Vacaville - Alamo	214	5.1%	0	65.0%	29.9%	0	0	0
Chula Vista - H St.	399	0	0.3%	97.0%	0.8%	0	2.0%	0
Modesto	227	9.7%	0	87.7%	0.4%	0	2.9%	0
Chula Vista - F St.	164	0	0	98.2%	1.8%	0	0	0
Ventura	87	0	0	62.1%	0	0	37.9%	0
La Mesa	36	8.3%	0	72.2%	8.3%	11.1%	0	0
Moorpark	60	0	28.3%	71.7%	0	0	0	0
RABO Total	2306	5.7%	0.9%	79.5%	6.1%	0.3%	7.4%	0.1%
Ontario	605	1.0%	0	79.7%	17.0%	0.3%	1.8%	0.2%
Highland	695	0.3%	0	97.6%	0.4%	0	1.7%	0
Non-RABO Total	1300	0.6%	0	89.2%	8.2%	0.2%	1.8%	0.1%
Total	3606	3.9%	0.6%	83.1%	6.8%	0.2%	5.4%	0.1%

¹ The most frequent combination for each row is bolded, and the second most frequent is italicized and underlined.

Although participants drove alone a majority of the time at each site, the proportion of drive-alone occasions varied from 57.7% to 98.2%, and each site had a somewhat different mode distribution. Other popular modes include walking/biking at Coronado (24.5%) and Davis (26.7%), carpooling at Alamo (29.9%) and Anaheim (21.9%), alternative-fuel vehicle (28.3%) at Moorpark Community College, and being dropped off at Grass Valley (23.2%) and Ventura Community College (37.9%).

However, it should be understood that the observed patterns are a function of the telecommuting frequency of individuals and the total number of individuals at each site. For example, two telecommuters were using the telecenter located at Moorpark Community College, and one of them commuted to the telecenter in an alternative-fuel vehicle. If this telecommuter were to telecommute more often, a higher utilization of this mode would result. This result does not imply that more people utilized that mode. Therefore, the mode distribution by sites only conveys information on what modes the telecommuters used and how many telecommuting occasions were made by that mode. The mode choice behavior of the individual telecommuter is discussed in Section 4.4.4.

4.4 Disaggregate Analysis

In the aggregate analysis, average telecommuting patterns for each center and across all telecenters were presented. The patterns are based on the pool of all telecommuting occasions considered together. Those who telecommute more often will have a disproportionate effect on the overall pattern, as their occasions appear more frequently in the pool. As a result, the aggregate patterns may not represent the behavior of an average telecommuter. To gain a better understanding of individual telecommuting behavior, the analysis reported in this section was performed on a person-by-person basis.

The study of telecommuting duration and frequency is fundamentally important to our understanding of the adoption of telecommuting and, hence, the impacts of telecommuting on travel and related issues. We may successfully predict that a certain number of individuals will telecommute. But if we falsely assume that they will telecommute in perpetuity (when in fact they, say, telecommute in a one-year-on, two-year-off cycle), and/or if we assume that they will telecommute (hypothetically) one day per week when the average is close to once every two weeks, we will greatly overestimate the number of people telecommuting on any given day and the associated travel-related impacts. The attendance log data collected for this study offers a unique opportunity to study these important questions, as well as other aspects of the telecommuting patterns of individuals.

The five disaggregate indicators analyzed here include individual telecommuting duration, telecommuting frequency, proportion of telecenter-only working days, work time spent at the telecenter, and proportion of drive-alone telecommuting days. However, not every individual in the data set was appropriate to include in the individual analyses. For 24 (20.7%) of the individuals at RABO sites and 45 (25.7%) at non-RABO sites, frequency and duration could not be meaningfully computed. These participants either telecommuted

• once only (13.8% at RABO sites; 20.6% at non-RABO sites),
• twice only, with less than two weeks between the two occasions (6.9% at RABO sites; 5.1% at non-RABO sites), or
• twice only, with more than a year between the two occasions (one user at a non-RABO site).

Some of these 69 individuals were new entrants to the program who had not had a chance to telecommute more often by the June 30, 1995 cutoff date, and others may have been drop-in users who were not expected to be regular telecommuters. However, at least 12 of these people were registered program participants who dropped out after one or two telecommuting occasions (participants who quit the program are discussed further in Chapter 5).

These 69 one- and two-time telecommuters were excluded from all disaggregate analyses. (Other participants who only telecommuted twice were retained and will appear with relatively short duration and/or frequency in the subsequent sections). The remaining 92 telecommuters at RABO sites and 130 individuals at non-RABO sites comprise the sample population of 2365 telecommuting occasions for the disaggregate analysis of telecommuting duration and frequency.

For the analysis of the individual proportion of telecenter-only working days, proportion of drive-alone telecommuting days, and work time, the information is based on the 92 telecommuters from RABO sites only. Non-RABO telecommuters are excluded because information was unavailable for many cases. The information on travel mode and work time distribution at multiple locations was unavailable for 70 (53.9%) of the non-RABO users because they stopped telecommuting prior to February 1, 1994 at Highland and September 1, 1994 at Ontario, the dates on which the attendance log designed by this project was introduced (see Appendix G). Only 25 (19.2%) non-RABO telecommuters had provided complete information on the five indicators listed above, and the other 35 (26.9%) individuals had provided partial information.

4.4.1 Telecommuting Duration

This section analyzes the length of time over which individuals telecommute from a center, or their telecommuting duration. We assume that there is no missing attendance information, that is, that each telecommuter signed in properly on each telecommuting day. For RABO sites, administrators had a vested interest in ensuring completeness of the sign-in records, as they were contractually obligated to maintain certain occupancy targets (see the companion volume on telecenter operation). Thus, while there are doubtless some missing data, it is reasonable to believe that they constitute a relatively small proportion of the whole. For the purposes of calculating telecommuting duration, it is assumed that no left-censoring of the data occurs, that is, that the individual had not been telecommuting prior to the first recorded use of the center. In other words, it is assumed that the first use of the center coincides with the start of telecommuting.

There is a difficulty, however, in similarly assuming that the last attendance date is the day the individual quit telecommuting. If this were the case, all telecommuters would be considered to have quit using the telecenter on or before June 30, 1995 (the last day in the interim report data set). In reality, of course, some participants will have quit before June 30 and others will not have. Since most of the participants did not telecommute every working day, or even with a constant frequency, it is difficult to determine whether a telecommuter had quit the program or was simply between uses of the telecenter.

Two decision rules were utilized to identify the status of the telecommuters as either quitters or stayers: one based on the existence of an exit interview and the other based on average length of time between telecommuting occasions. The telecommuters who were known to have stopped telecommuting were asked to participate in an exit interview as a part of this project. Those who completed an exit interview were easily identified as quitters (see Chapter 5). For the rest of the telecenter users, a rule was devised to define their telecommuting status. If the period of time from the last telecommuting date to the cutoff date for inclusion in the interim report (June 30, 1995) was more than three times the average length of time between two successive telecommuting occasions for that person, the telecommuter was regarded as a quitter. Otherwise, s/he was a stayer, meaning that the actual exit time-point had not yet been observed for that individual. Thus, the telecommuting durations of stayers are right-censored. Although arbitrary, using three time the average period between telecommuting occasions as the basis for a decision rule is based on the concern that the telecenter users may reduce their telecommuting frequency but still remain in the program. Nevertheless, applying this rule runs some risk of falsely classifying as stayers people who quit telecommuting shortly before June 30, as well as a risk of misclassification in the opposite direction.

Therefore, the definition of telecommuting duration is based on whether the telecommuter is a quitter or stayer. For quitters, the last day of telecommuting is considered to be the date of their final attendance log entry. However, stayers are considered to be telecommuting up to the cutoff date of June 30, 1995 instead of up to the last recorded day of telecommuting. For example, if a stayer's last recorded telecommuting occasion before the cutoff date was on June 21, 1995, the duration is counted from the first telecommuting date to June 30, 1995. In addition, the duration is rounded down to the nearest month. For example, if an individual telecommuted for 3.8 months, s/he is classified as a stayer for the first three months and as a quitter for the fourth month.

Telecommuting duration here is similar to the survival time of an individual in a conventional medical study: those who quit telecommuting are analogous to the patients who die and the stayers are analogous to those who are living at the end of the observation period. The data possess two features which correspond to the characteristics of survival data. First, telecommuting duration is not symmetrically distributed: some telecenter users quit within a very short time but some continue to telecommute for more than three years (at non-RABO sites). Second, as discussed above, the telecommuting duration is frequently right-censored.

The ratio between quitters and stayers is of importance to the analysis. If the right-censored observations (stayers) outnumber the uncensored ones (quitters), the statistical techniques associated with failure time data may not be appropriate for this study. Of the 222 individuals (92 at RABO sites and 130 at non-RABO sites) considered, 47 of the RABO telecenter users (51.1%) and 99 non-RABO users (76.2%) stopped telecommuting on or before the June 30, 1995 cutoff date. This proportion of quitters is considered acceptable for the use of conventional failure time analysis techniques. The following is a summary of the model formulation drawn from Miller (1981), Cox and Oakes (1984), and Collett (1994).

A basic element in the analysis of telecommuting duration is the survival function. The survival function is defined as the proportion of telecenter users telecommuting beyond time t:

= (# of telecenter users telecommuting longer than t months)/(total # of telecenter users) (4.3)

Suppose that there are n telecommuters for whom telecommuting durations are observed. Some of these observations are right-censored, and there is also more than one telecommuter with the same observed exit time. Therefore, suppose there are r exit times among the individuals, where r .LE. n. Then these exit times are arranged in ascending order: t(1) < t(2) < ... < t(r). The probability of surviving at a specific time t(j) given that the individual has already survived past time t(j - 1) could be estimated as

(4.4)

where T is the observed telecommuting duration, n_j is the number of individuals who were still telecommuting just before t(j) and q_j is the number of individuals who quit in the time interval [t(j), t(j + 1)).

The number of telecommuters n_j is governed by the equation

n _j = n _{j - 1} - q _{j - 1} - c _{j - 1} ,              (4.5)

where c_j - 1 is the number of censored observations in the time interval [t(j), t(j + 1)). The status of observations that are censored at time t(j - 1) cannot be determined for later times, and hence these censored observations must be removed from the number of people n_j known to be telecommuting at times t(j) and later.

Suppose the exit times of telecommuters are assumed to occur independently. A series of time intervals can be constructed based on the observed exit times of the telecommuters. The cumulative probability of surviving beyond the kth exit time is the product of these k interval-specific survival probabilities:

(4.6)

Tables 4-8 and 4-9 illustrate the estimated survival functions for the telecommuters at RABO sites and non-RABO sites, respectively. These functions indicate the probabilities that an individual continues to telecommute after each time interval. From Table 4-8, for example, the probability of telecommuting beyond six months (through the six intervals) is

(4.7)

Telecommuting Duration (months) j	Initial Number nj	Number of Quitters qj	Number of Censored Observations cj	Conditional Probability of Surviving Beyond t(j) P(t(j))	Cumulative Probability of Surviving Beyond t(j)
0 - 1	130	0	0	1.000	1.000
1 - 2	130	6	0	0.954	0.954
2 - 3	124	11	1	0.911	0.869
3 - 4	112	5	1	0.955	0.830
4 - 5	106	10	0	0.906	0.752
5 - 6	96	13	2	0.865	0.650
6 - 7	81	6	0	0.926	0.602
7 - 8	75	4	2	0.947	0.570
8 - 9	69	5	1	0.928	0.529
9 - 10	63	6	1	0.905	0.478
10 - 11	56	4	2	0.929	0.444
11 - 12	50	3	0	0.940	0.418
12 - 15	47	2	4	0.957	0.400
15 - 16	41	1	1	0.976	0.390
16 - 18	39	3	1	0.923	0.360
18 - 19	35	2	0	0.943	0.339
19 - 20	33	2	0	0.939	0.319
20 - 21	31	1	0	0.968	0.309
21 - 23	30	2	0	0.933	0.288
23 - 24	28	2	0	0.929	0.267
24 - 25	26	2	1	0.923	0.247
25 - 26	23	2	2	0.913	0.225
26 - 31	19	2	4	0.895	0.202
31 - 32	13	1	2	0.923	0.186

Telecommuting Duration (months) j	Initial Number nj	Number of Quitters qj	Number of Censored Observations cj	Conditional Probability of Surviving Beyond t(j) P(t(j))	Cumulative Probability of Surviving Beyond t(j)
32 - 40	10	1	3	0.900	0.168
40 - 41	6	1	0	0.833	0.140
41 - 42	5	1	4	0.800	0.112
42+	4	1	3	0.750	0.084

That is, there is a 54.5% chance that an individual will telecommute longer than six months. In a similar way, we can obtain a 27.4% chance of telecommuting more than 15 months. From the P(t(j)) column of Table 4-8 it is seen, for example, that there is a 91.7% chance of continuing to telecommute past six months given that the individual has lasted five months. The graph of the estimated survival functions from BMDP software is shown in Figure 4-8. The estimated survival functions are constant between adjacent exit times and decrease at each exit time.

The median duration of telecommuting was 8 months at RABO sites and 9 months at non-RABO sites. This means that 50% of the participants telecommuted at least 8 and 9 months, respectively. Put negatively, it also means that half of the participants telecommuted at most 8 or 9 months. More than 25% of the RABO telecommuters telecommuted for at least a year, compared to 40% of non-RABO users. About 25% of the non-RABO telecommuters used the telecenter for at least two years. Despite the difference in telecommuting duration between RABO and non-RABO individuals, the two survival functions were not statistically different at a 0.10 level of significance. This means that at any time t, the estimated survival probability of telecommuting beyond t is statistically the same for telecommuters at both RABO and non-RABO sites. This result suggests that the operating length of the telecenter may not be an important factor in determining telecommuting duration. Rather, duration is probably a function of the characteristics of the individual telecommuter.

This relatively short median duration of telecommuting is an important finding. Few studies have collected data on attrition in telecommuting, so there is little to which to compare this figure. However, one study of home-based telecommuting reported an attrition rate of 33% within one year (Quaid and Lagerberg, 1992). Thus, these two studies suggest that attrition is higher for center-based telecommuting than for the home-based form, but further research is needed on this point.

Based on the analysis in this chapter, "once a telecommuter, always a telecommuter" is clearly not true. Reasons for quitting telecommuting are discussed in Chapter 5. In any case, later discussions of telecommuting frequency (Section 4.4.2) and of the travel impacts of telecommuting (Chapter 6) should be interpreted in the light of this information: that is, measured telecommuting frequencies and impacts may only be achieved for a relatively short period of time.

4.4.2 Telecommuting Frequency

In order to measure how often telecommuters used the telecenter, an individual's average telecommuting frequency is taken to be the ratio of the number of telecommuting days to the total number of working days during the duration of telecommuting. Again, this assumes no missing telecommuting occasions for each telecommuter. The number of working days includes the first and last telecenter visits but excludes Saturdays, Sundays, and eight federal holidays (see Section 4.3.4).

The frequency calculations were slightly modified to accommodate missing data at non-RABO sites. Data was missing for January 1994 at Highland and August 1993 at Ontario. The first and last telecommuting occasions could be used to judge whether the telecommuting duration included a missing month, but no information existed on specific telecommuting occasions. According to data from other months, 30 telecommuters may have telecommuted in those two months. For these cases, the number of working days in the missing month was subtracted from the total number of working days. Unless the telecommuting frequency for an individual was much higher or much lower than average during that month, the estimated frequency should be reliable.

Figure 4-9 shows the distribution of the average frequency of telecommuting for the 92 telecommuters at RABO sites and the 130 telecommuters at non-RABO sites. The cumulative distribution is shown in Figure 4-10. Since there are about 21 working days per month on average, a 5% telecommuting frequency is approximately equivalent to one telecommuting day per month. A 20% telecommuting frequency represents telecommuting once per week and 40% means twice per week. At RABO sites, about 9% of the telecommuters telecommuted on fewer than 5% of their working days. This implies that, for them, the average length of time between two telecommuting occasions was more than a month. More than half of the RABO telecommuters telecommuted less than one day per week, and about 22% telecommuted one to two days per week, on average.

The average telecommuting frequency at non-RABO sites (17.2%) was lower than that at RABO sites (25.0%). Nearly 21% telecommuted less than once per month on average. About 75% telecommuted less than one day per week. Although the low average frequency may reflect the real behavior of the telecommuters, the longer average telecommuting duration may be related to the low frequency. The longer period of observation at non-RABO sites may include a period of no telecommuting by the participants since some of the users were found to stop telecommuting for an extended period of time and then restart later on. Another possible explanation of the difference is that non-RABO site users may have been more likely not to sign in on occasions when they actually did use the center. Since non-RABO sites did not have the same contractual obligation to maintain target occupancy levels as the RABO sites did (see the companion volume on telecenter operation), they may not have rigorously enforced a policy of signing the attendance log on every occasion.

From the presentation of the average telecommuting frequency, it should not be inferred that telecommuters had a constant telecommuting frequency. The telecommuters are likely to have had several periods with different telecommuting frequencies during the entire duration of telecommuting. Therefore, the average frequency only reflects aggregate individual telecommuting behavior.

4.4.3 Comparison of Different Measures of Telecommuting Frequency

The center-based telecommuting frequency for project participants may be estimated from two sources, namely, the after attitudinal surveys and the sign-in logs. The preceding section discussed the distribution of telecommuting frequencies based on the complete available sign-in log data for 92 RABO and 130 non-RABO telecommuters. Average frequencies of 25.0% for RABO sites and 17.2% for non-RABO sites were found, for telecommuting durations ranging from one to forty-three months. However, from the attitudinal survey data for the 39 RABO and non-RABO respondents who completed after surveys (Table 3-10 in Section 3.2.4), a "current" average telecommuting frequency of 29.9% was computed. This difference between data sources may be due to differences in the sample (those who completed the after survey may have been higher-frequency telecommuters), changes in the frequency of telecommuting over time, and/or a survey response bias.

We chose to explore further the third possibility, that of a survey response bias. In particular, it is of interest to obtain some insight into how respondents interpreted the attitudinal survey question (D11a, see Appendix E), "How much do you currently telecommute from a telecommuting center?" Since no specific time frame was given in the question, several interpretations are plausible. Respondents may have tended to report their most recent frequency (say, over the last month), an average frequency since the start of telecommuting, or some perceived "typical" frequency which may or may not relate to either of the previous possibilities. It is hypothesized that responses to the attitudinal survey will tend to overstate the actual amount of telecommuting. There may be a number of reasons for this, including the tendency to telescope less frequent events into a shorter time frame than the actual, a desire to increase the apparent success of the program, and "wishful thinking" ­ that is, a tendency to confound the actual frequency of telecommuting with a desired, perhaps an explicitly-stated, target frequency. To examine this hypothesis, the sign-in log data for the 39 attitudinal survey respondents was used to obtain the telecommuting frequency both during a one-month and a six-month window prior to the date on which the respondents filled out the after attitudinal surveys.

For the purposes of comparing these alternate measures of telecommuting frequency, a month was considered to have 22 working days. Holidays are disregarded, which means that the aggregate results discussed here slightly underestimate the frequency of telecommuting as a proportion of actual workdays. However, in reporting their telecommuting frequency as a general rate (for example, 1 to 2 days per week) in the attitudinal survey, it is unlikely that respondents precisely factored in the influence of holidays. In any case, the assumption is a convenient simplification, and as it is applied to all frequency measures equally, it should not affect the results of the comparison. Also, in calculating the average frequency from the attitudinal survey, the midpoint of the category checked was initially taken as the telecommuting frequency for that person (for example, the response category "about 1-3 days a month" was treated as a telecommuting frequency of 2/22, or 9.09%). The average frequencies obtained from the 39 attitudinal surveys and the corresponding sign-in log entries are shown below in the first three rows of Table 4-10.

As hypothesized, the highest measure of telecommuting frequency is obtained from the attitudinal survey: 30%, or 1½ days per week on average. The next highest measure is the one-month window from the sign-in log, showing an average 20% or one-day-per-week frequency. The six-month sign-in log measure is 18%, or 0.9 days per week. From a comparison of the two sign-in log averages it may appear that the telecommuting frequency is increasing over time, but the difference is not statistically significant (t = 0.47; p-value = 0.644). However, the differences between AS and SIL6 (t = 2.16; p-value = 0.040), and between AS and SIL1 (t = 1.68; p-value = 0.100) are statistically significant (at a .LE. 0.10). Thus, initially the evidence seems to support the hypothesis that on the attitudinal survey, respondents overstate their actual telecommuting frequency as determined by the attendance log.

Table 4-10: Average Telecommuting Frequency

¹ Using the lower bound rather than the midpoint of each category checked as the telecommuting frequency.

However, another potential cause for this result should be examined. It may be that the majority of the frequencies reported in the attitudinal survey actually fell in the lower half of the chosen categories. For example, in the "1-2 days a week" category, it is possible (even likely) that more respondents telecommute one day per week than two days per week. If this were true, then taking the midpoint of the interval as the average frequency for each category artificially inflated the overall average.

To further examine this potential cause of the observed results, the lower bound rather than the center of the interval was taken as the representative value for each category in the attitudinal survey, and the telecommuting frequency average was re-calculated. Now, if the obtained average were still significantly different from the averages obtained from SIL1 and SIL6, then a reasonable conclusion would be that the respondents consistently over-reported their frequencies. However, if the obtained average were not significantly different, then either or both of the above two reasons could be valid. Note that using the lower bound as the representative value is a conservative test, as the true average for the category is almost certainly higher than the lower bound.

The average telecommuting frequency obtained from the attitudinal survey, using the lower bound rather than the midpoint of the interval as the frequency value for each category, is presented in the last row of Table 4-10. Using this approach, the average obtained from the attitudinal surveys is still greater than the averages obtained from the one-month and six-month sign-in log data. However, t-tests reveal that the differences between AS and SIL6 (t = 1.01; p-value = 0.328) and between AS and SIL1 (t = 0.56; p-value = 0.582) are not statistically significant. This implies that the higher average obtained from the attitudinal surveys could be the result of either an over-reporting bias on the part of the respondents in the attitudinal survey, or an artifact in the way point frequencies were estimated from the categorical (interval) responses.

This equivocal result calls for even deeper exploration. Since we have the sign-in log data representing actual telecommuting frequencies, we can reconstruct the true distribution of frequencies across the sample and compare that to the distribution based on the reported frequencies of the attitudinal survey. If the observed result is due to the fact that actual frequencies tend to fall in the lower half of the reported frequency category, the distributions from the two sources will match relatively closely at the category level. If, on the other hand, reported frequencies tend to overstate actual frequencies, then there will be a mismatch of the two distributions, with the reported frequency distribution disproportionately skewed toward higher frequency categories.

Figure 4-11 and Table 4-11 show the telecommuting frequency distribution from the attitudinal survey data, six-month sign-in log data, and the one-month sign-in log data. Visually, the figure indicates that although the AS distribution is relatively similar to the SIL1 distribution, it is skewed slightly upward. In particular, several more respondents reported an AS frequency of three or more days per week than actually telecommuted that often within the last month, which accounts for the higher average frequency obtained from the attitudinal survey. Nevertheless, a chi-squared test emphatically fails to reject the hypothesis that the two distributions are equivalent. However, the differences between AS and SIL6 are more pronounced (especially in the same three-or-more-days-per-week range), with a chi-squared test rejecting the equivalency hypothesis at a level of significance of 0.10 but not at 0.05.

SIL6-AS : c² = 6.52 (critical c² (3, 0.05) = 7.81)
SIL1-AS : c² = 1.52 (critical c² (3, 0.05) = 7.81)
(The c² values were calculated after combining 5/week cells with 3-4/week cells. )

In the above table, the two numbers in the parentheses depict the number of respondents whose frequencies fell in the lower half and the upper half of the category, respectively. From the figures in the parentheses, we see that for SIL6 the frequency values do not predominantly lie in the lower half of the categories (16 lower, 23 upper), whereas for SIL1 they do (27 lower, 12 upper).

These results indicate that the telecommuting frequencies reported in the attitudinal survey do not differ significantly from the actual frequencies within the preceding month. There is a statistically weak, but suggestive, indication that the reported frequencies are higher than the actual frequencies over the preceding six months. The implication is that participants most heavily base their reported frequency on their most recent frequency (over the last month) rather than on their average frequency over a period of several months.

4.4.4 Proportion of Telecenter-only Working Days

In the aggregate analysis, it was found that the RABO telecommuters worked entirely at the telecenter for nearly 60% of the days on which telecommuting occurred. Individual telecommuting behavior is further analyzed below.

Figure 4-12 illustrates the distribution of the proportion of telecommuters that worked entirely at the telecenter on telecommuting days. The ratio is calculated by dividing the number of telecommuting-only days over the total number of telecommuting occasions for an individual. Among the 92 telecommuters at RABO sites, 21.8% worked entirely at the telecenter on all telecommuting days, and another 22.8% had a high proportion (80%-99%) of telecenter-only days. It was also found that seven telecommuters (7.6%) always worked at additional workplaces on all telecommuting days. Approximately 34% of the RABO telecommuters worked at more than one location on telecommuting days at least 60% of the time.

The distribution according to each site shown in Table 4-12 has some consistency with the patterns found in the aggregate analysis (shown in Table 4-5). For the sites with a large proportion of telecenter-only telecommuting occasions, such as at Modesto, 90% of the telecommuters worked only at the telecenter on their telecommuting days. For those sites with a small proportion of telecenter-only occasions, such as Ulatis, Alamo, and Chula Vista (H St.), about half of the telecommuters had telecenter-only occasions 40% of the time or less. The implication is that these individuals frequently worked at two or more workplaces on telecommuting days. For the sites with few telecommuters, such as Moorpark Community College and Ventura Community College, the distribution shown in Table 4-12 indicates that some site-specific patterns shown in Table 4-5 are heavily influenced by one or two individuals.

4.4.5 Individual Work Time Spent at the Telecenter

From the aggregate analysis, it was found that work time spent at the telecenter on telecommuting occasions varied because telecommuters worked at a number of different locations in the same day. In this section, an attempt is made to establish the pattern of individuals. The behavior at a disaggregate level may differ from the average results of the site-level analysis.

The average work time spent at telecenters by an individual was calculated by dividing the sum of work time for all his/her telecommuting occasions by the total number of those occasions (missing data was excluded). The distribution of the resulting average work time is presented in Figure 4-13. Nearly 53% of the telecommuters worked at the telecenters more than six hours per telecommuting occasion on average. The majority (71.7%) stayed at the telecenters for more than four hours. This shows that the telecenter was the main workplace for most of the telecommuters on telecommuting days even though they might have more than one work location. A substantial minority (more than a quarter of the sample), however, typically used the telecen ter for half a day or less ­ either as a drop-in location between work-related meetings elsewhere or in conjunction with telecommuting from home or commuting to the conventional workplace.

Table 4-12: Distribution of the Percentage of Telecenter-only Working Days by RABO Telecommuters

Site	N1	0%	1-20%	21-40%	41-60%	61-80%	81-99%	100%
Coronado	9	22.2%	22.2%	---	11.1%	11.1%	33.3%	---
Grass Valley	8	---	12.5%	---	---	---	50.0%	37.5%
Anaheim	10	---	10.0%	10.0%	20.0%	10.0%	40.0%	10.0%
Vacaville (2 Sites)	26	3.8%	11.4%	38.6%	15.4%	7.7%	15.4%	7.7%
Davis	3	---	---	---	33.3%	33.3%	---	33.3%
Chula Vista - H St.	12	16.7%	16.7%	16.7%	8.3%	---	25.0%	16.7%
Modesto	10	10.0%	---	---	---	---	---	90.0%
Chula Vista - F St.	5	20.0%	---	---	20.0%	20.0%	20.0%	20.0%
Ventura	3	---	33.3%	33.3%	---	---	---	33.3%
La Mesa	4	---	---	---	---	25.0%	50.0%	25.0%
Moorpark	2	---	---	---	100%	---	---	---
Total	92	7.6%	12.0%	14.1%	13.0%	8.7%	22.8%	21.8%

¹N is the number of telecommuters.

Interestingly, a significant proportion of the RABO telecommuters (8.7%) were likely to work more than eight hours per day at the telecenter. On average, 41.2% worked at the telecenter for at least seven hours per telecommuting day. However, from the distribution shown in Figure 4-12, only 21.8% worked exclusively at the telecenter. If they all worked for at least seven hours, there still is a significant proportion of individuals (19.4%) who not only worked at the telecenter for at least seven hours on average but also spent some time at other workplaces.

4.4.6 Mode Choice to the Telecenter

The examination of the mode choice behavior of the telecommuters uses an approach similar to the previous analysis of individual patterns of workplace use. The frequency of driving alone on telecommuting days is calculated through dividing the number of telecommuting occasions on which a person drove alone by the total number of telecommuting occasions for that individual.

Figure 4-14 shows the corresponding distribution at RABO sites. About 46% of the telecommuters drove alone to the telecenter for all telecommuting occasions. More than two-thirds (71.2%) drove alone frequently (more than 75% of their occasions). About 54% used other modes, such as carpooling or being dropped off, to reach the telecenter at least occasionally. Only 4% of the telecommuters never drove alone to the center. The above findings confirm that driving alone was the prevailing transportation mode of choice despite the effort to locate the centers close enough to residential neighborhoods so that walking and biking would be attractive options.

4.5 Summary and Conclusions

This chapter describes a study of the telecommuting patterns of center-based telecommuters, taken primarily from information compiled from the attendance logs at the telecenters. This analysis identifies patterns of telecommuting duration and frequency, and increases our understanding of telecommuter working behavior on telecommuting days. The study analyzes telecommuting patterns both at the aggregate (site) level and the disaggregate (individual) level. Both analyses are consistent with each other and complementary.

For most of the telecenters, a usage rate of between 10% and 20% was maintained. Though the usage rates fluctuated, overall growth is apparent. As of the June 30, 1995 cutoff date for this interim report, the RABO telecenters had been open an average of 9.1 months, with a minimum of 2.5 months and a maximum of a little more than 20 months. The two non-RABO telecenters have been operating for much longer, an average of 3.1 years.

At RABO sites, the average telecommuting frequency was 25%, or 1¼ days per week. More than half of the telecommuters telecommuted less than one day per week on average, and 22% telecommuted 1 to 2 days per week. The non-RABO telecommuters telecommuted less frequently than those who were at RABO sites; the average was 17.2%, with about 75% of non-RABO telecenter users telecommuting less than one day per week.

Attrition at the telecenters was relatively high, with 50% of all telecommuters quitting within the first nine months. Although little comparative data are available, this appears to be higher than for home-based programs. Reasons for quitting telecommuting are analyzed in Chapter 5. But in any case, the frequency and distribution of telecommuting are crucial factors to consider in any forecast of levels and impacts of telecommuting. Of the 92 RABO participants who telecommuted often enough to analyze, half telecommuted for at least 8 months, and more than 25% telecommuted for at least one year. At non-RABO sites, 50% of the 130 telecommuters analyzed telecommuted for at least 9 months, and 25% telecommuted for at least 2 years. There is no significant difference in telecommuting duration between RABO and non-RABO sites.

A majority of telecommuters (53%) worked at the telecenters for at least 6 hours on average on their telecommuting days. The most common telecommuting pattern was to work entirely at the telecenter. Approximately 22% of the telecommuters at RABO sites telecommuted with this pattern on all of their telecommuting occasions, and an additional 23% did so at least 80% of the time. At least 34% usually worked at more than one work location, including 8% who always did. The second most common workplace combination was telecenter/other work location (i.e., other than home or the regular workplace). Contrary to expectation, center- and home-based telecommuting are not often combined on the same day; patterns involving these two locations occurred only 17% of the time at RABO sites.

Driving alone was the dominant transportation mode used by the telecommuters in commuting to the center. About 46% of the RABO telecommuters drove alone to the center on all of their telecommuting occasions. More than two-thirds drove alone to the center very frequently (more than 75% of their occasions).

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.