Tải bản đầy đủ - 0 (trang)
1 Descriptive Statistics for the Household- and MSA-Level Dependent and Independent Variables

# 1 Descriptive Statistics for the Household- and MSA-Level Dependent and Independent Variables

Tải bản đầy đủ - 0trang

Multilevel Models of Commute Times for Men and Women

203

all regions. The largest gender gap was in the Northeast (4:58 min) and the smallest

gender gap was in the West (1:51 min). Regional comparison of commute times

was examined by estimating the coefficient of variation, expressed as a percentage,

nationally and for each regional subsample. Nationally, coefficients of variation in

commute times for men (66.69%) and women (67.19%) were approximately the

same. Regionally, however, coefficients of variation in commute times were lower

for men than women in the Northeast (69.72% vs. 72.05%) and West (66.04%

vs. 69.17%) and higher in the Midwest (65.28% vs. 60.78%) and South (63.73%

vs. 63.27%). The detected regional differences in commuting behavior support the

argument to control for regional variation in private-vehicle commute times in the

multilevel models.

Table 1 also provides descriptive statistics for the household-level independent

variables for all subsamples. The descriptive statistics for continuous independent

variables are the mean and standard deviation, while the descriptive statistic for

discrete independent variables is the percentage. Except for income, the categories

with the highest percentages in the men-only, women-only, and pooled men–women

subsamples were approximately the same. That is, the income category with the

highest percentage of men was \$50,000 to \$74,999 (25.60%), while the income

category with the highest percentage of women was \$25,000 to \$49,999 (30.27%).

Such a result reaffirms the notion that women earn less than men. On the one

hand, the typical respondents in the men-only, women-only, and pooled men–

women subsamples were:

43 years old;

white;

married/partnered with children; and

employed full-time in professional occupations.

Consistent with empirical evidence on gender differences in access to private vehicles, the typical male respondent lived in a household with a 1.31 vehicle to worker

ratio, while the typical female respondent lived in a household with a 1.14 vehicle

to worker ratio. The other household characteristics that most distinguished men

and women in the subsamples, besides income and the ratio of vehicles to workers

were occupation and employment status (full-time or part-time). As expected, more

women were employed in clerical occupations, while more men were employed

in manufacturing occupations. Also as expected, more men worked full-time and

more women worked part-time. Overall, however, these descriptive data lend support to the hypothesis that the socioeconomic characteristics of men and women are

presently more alike than in the past.

Table 2 provides descriptive statistics for the MSA-level independent variables.

The typical MSA:

had a congestion measure of 1.22, 30:00-min trips during free-flow periods took

37:00 min during peak periods;

had a land area of 1,746.26 km2 ;

had a population size greater than or equal to 3,000,000;

204

E.J. Zolnik

Table 1 Descriptive statistics for household-level dependent and independent variables for menonly, women-only, and pooled men–women subsamples

Variable

Category

Men

Women

Men–Women

Commute time (min)

25:91.17:28/

22:80.15:32/

25:35.16:40/

Age

43:25.11:35/

42:99.11:11/

43:12.11:23/

Ethnicity

White

African American

Asian

AI/ANa

NH/PIb

Hispanic/Mexican

Other

78:16%

5:76%

6:56%

0:27%

0:80%

3:24%

5:21%

77:36%

8:80%

4:34%

0:60%

0:82%

2:99%

5:06%

77:76%

7:28%

5:45%

0:44%

0:81%

3:12%

5:15%

Income

<\$25,000

\$25,000 to \$49,999

\$50,000 to \$74,999

\$75,000 to \$99,999

\$100,000

6:13%

25:18%

25:60%

20:17%

22:91%

9:37%

30:27%

23:59%

18:05%

18:72%

7:75%

27:72%

24:59%

19:11%

20:82%

Life cycle

18:72%

2:79%

35:18%

43:31%

18:85%

9:60%

31:26%

40:29%

18:79%

6:20%

33:22%

41:80%

Occupation

Service

Clerical

Manufacturing

Professional

20:62%

3:64%

22:69%

53:05%

20:67%

24:41%

4:14%

50:79%

20:64%

14:02%

13:41%

51:92%

Work

Full-time

Part-time

Multiple jobs

Vehicles to workers

94:56%

4:94%

0:50%

1:31.0:65/

80:88%

18:67%

0:45%

1:14.0:45/

87:72%

11:81%

0:47%

1:22.0:57/

Descriptive statistics reflect means and standard deviations for continuous variables and percentages for discrete variables. The standard deviations appear in parentheses after the means for the

continuous variables and the percentages refer to the share of the subsample in each category of

the discrete variables.

a

b

Native Hawaiian/Pacific Islander.

was located in the Southern region of the coterminous US;

had a residential density score of 104.37, a land use mix score of 99.97, a degree

of centering score of 101.59, and a street accessibility score of 102.12; and

Multilevel Models of Commute Times for Men and Women

205

Table 2 Descriptive statistics for MSA-level independent variables for men-only, women-only,

and pooled men–women subsamples

Variable

Category

Men

Women

Men–Women

Congestion

Travel time index

1:22.0:10/

1:22.0:10/

1:22.0:10/

1:75.1:35/

1:75.1:35/

1:75.1:35/

Land area (1,000 km2 /

Population size

Medium (0.5M to <1M)

Large ( 1M to 3M)

Very large ( 3M)

6:68%

31:71%

61:61%

5:98%

34:51%

59:51%

6:33%

33:11%

60:56%

Region

Northeast

Midwest

South

West

29:69%

17:73%

30:62%

21:96%

29:97%

19:20%

30:82%

20:02%

29:83%

18:46%

30:72%

20:99%

Sprawl

Residential density

Land use mix

Degree of centering

Street accessibility

Commuter rail (Yes)

Commuter rail (No)

104:37.27:47/

99:97.20:04/

101:59.22:17/

102:12.26:22/

50:91%

49:09%

104:37.27:47/

99:97.20:04/

101:59.22:17/

102:12.26:22/

50:69%

49:31%

104:37.27:47/

99:97.20:04/

101:59.22:17/

102:12.26:22/

50:80%

49:20%

Descriptive statistics reflect means and standard deviations for continuous variables and percentages for discrete variables. The standard deviations appear in parentheses after the means for the

continuous variables and the percentages refer to the share of the subsample in each category of

the discrete variables.

5.2 Men-Only Multilevel Model

Results from the household-level of the men-only multilevel model appear in the

Men column of Table 3. Only statistically significant coefficients are reported. The

referent category for each discrete independent variable represents the typical male

respondent. At the household-level, men whose total income was less than \$25,000

and \$25,000 to \$49,999 commuted 4:02 and 3:35 min less, respectively, than men

whose total income was \$50,000 to \$74,999. Single men with no children commuted

2:37 min less than married/partnered men with children. Men who worked part-time

commuted 3:35 min less than men who worked full-time. Finally, a one unit increase

in the vehicle to worker ratio increased commute times for men by 1:20 min.

Results from the MSA-level of the men-only multilevel model appear in the Men

column of Table 4. Only statistically-significant coefficients are reported. The referent category for the discrete independent variables population size, region, and

commuter rail represents the typical MSA. At the MSA-level, a one unit increase

in the value of the congestion measure and land area is associated with an increase

in the commute times of men of 7:07 and 0:01 min, respectively. Commute times

206

E.J. Zolnik

Table 3 Household-level coefficients and standard errors for men-only, women-only, and pooled

men–women multilevel models

Variable

Category

Men

Women

Men–Women

Y-Intercept

27.69 (0.46)

26.96 (0.77)

2.48 (0.01)

Age

Ethnicity

White

Referent

Referent

Referent

African American

Asian

AI=ANa

NH=PIb

6.40 (2.93)

Hispanic/Mexican

Other

0.05 (0.02)

Income

2:19 (0.79)

0:05(0.03)

<\$25,000

4:03(1.13)

\$25,000 to \$49,999

3:59(0.99)

Referent

Referent

\$50,000 to \$74,999

Referent

0.05 (0.02)

\$75,000 to \$99,999

2.15 (0.70)

0.11 (0.02)

\$100,000

3.11 (0.47)

0.13 (0.02)

Life cycle

2:62(0.64)

Referent

Referent

Referent

Occupation

0:07(0.02)

Service

3:29(0.75)

Clerical

1:34(0.67)

0:05(0.02)

Manufacturing

1.94 (0.74)

Professional

Referent

Referent

Referent

Work

Full-time

Referent

Referent

Referent

3:01 (0.45)

0:15(0.01)

Part-time

3:59(0.48)

Multiple jobs

0.91 (0.50)

0.04 (0.01)

Vehicles to workers

1.33 (0.48)

Referent category represents typical respondent. Standard errors appear in parentheses after coefficients. , , and

indicate 90%, 95%, and 99% significance levels, respectively.

a

b

Native Hawaiian/Pacific Islander.

for men were, on average, 2:36 min shorter in large population-sized MSAs than in

very large population-sized MSAs. Finally, a one unit increase in residential density

score is associated with a decrease in commute times of men of 0:01 min. Overall,

results from the men-only multilevel model suggest that:

low-income men commute less than middle-income men;

single men without children commute less than married/partnered men with

children;

Multilevel Models of Commute Times for Men and Women

207

Table 4 MSA-level coefficients and standard errors for men-only, women-only, and pooled men–

women multilevel models

Variable

Category

Men

Women

Men–Women

Congestion

11:18 (4.76)

0.18 (0.07)

Travel time index

7.11 (3.18)

Land area .km2 /

0.001 (2e-4)

0.001 (3e-4)

0.00003 (5e-6)

Population size

Medium (0.5M to <1M)

0:08 (0.02)

Large ( 1M to <3M)

2.95 (0.67)

Very large ( 3M)

Referent

Referent

Referent

Region

Northeast

4:79 (0.03)

Midwest

3:66 (0.60)

South

Referent

Referent

Referent

West

Sprawl

0:04 (0.01)

0:001 (2e-4)

Residential density

0:02 (0.01)

Land use mix

0:001 (4e-4)

Degree of centering

Street accessibility

Commuter rail (Yes)

Referent

Referent

Referent

Commuter rail (No)

2:85 (0.59)

Referent category represents typical respondent. Standard errors appear in parentheses after

coefficients. , , and

indicate 90%, 95%, and 99% significance levels, respectively.

men who work part-time commute less than men who work full-time; and

men with more access to private vehicles commute more than men with less

On average, commute times in the men-only subsample were shorter in less congested, large population-sized MSAs.

5.3 Women-Only Multilevel Model

Results from the household-level of the women-only multilevel model appear in the

Women column of Table 3. Only statistically-significant coefficients are reported.

The referent category for each discrete independent variable represents the typical female respondent. At the household-level, Native Hawaiian/Pacific Islander

women commuted 6:24 min more than white women. Women whose total income

was less than \$25,000 commuted 2:11 min less than women whose total income was

\$25,000 to \$49,999, while women whose total income was \$75,000 to \$99,999 and

greater than or equal to \$100,000 commuted 2:09 and 3:07 min more, respectively,

than women whose total income was \$25,000 to \$49,999. Women employed in service and clerical occupations commuted 3:17 and 1:20 min less, respectively, than

208

E.J. Zolnik

women employed in professional occupations, while women employed in manufacturing occupations commuted 1:56 min more than women employed in professional

occupations. Women who worked part-time commuted 3:01 min less than women

who worked full-time. Finally, a one unit increase in the vehicle to worker ratio

increased commute times for women by 0:55 min.

Results from the MSA-level of the women-only multilevel model appear in the

Women column of Table 4. Only statistically-significant coefficients are reported.

The referent category for the discrete independent variables population size, region,

and commuter rail represents the typical MSA. At the MSA-level, a one unit

increase in the value of the congestion measure is associated with a decrease in

the commute times of women of 11:11 min, while a one unit increase in land area is

associated with an increase in the commute times of women of 0:01 min. Commute

times for women were, on average, 4:47 and 3:40 min shorter in Northeastern and

Midwestern MSAs, respectively, than in Southern MSAs. A one unit increase in residential density score is associated with a decrease in the commute times of women

of 0:02 min. Finally, commute times for women were 2:51 min shorter in MSAs that

had commuter rail than in MSAs that did not have commuter rail.

Overall, results from the women-only multilevel model suggest that:

Native Hawaiian/Pacific Islander women commute more than white women;

high-income women commute more than low-income women;

women with manufacturing jobs commute more than women with service jobs;

women who work part-time commute less than women who work full-time; and

women with more access to private vehicles commute more than women with

On average, commute times in the women-only subsample were shorter in more

congested, Northeastern and Midwestern MSAs that had commuter rail.

5.4 Pooled Men–Women Multilevel Model

Results from the household-level of the pooled men–women multilevel model

appear in the Men–Women column of Table 3. Only statistically significant coefficients are reported. The referent category for each discrete independent variable

represents the typical male and female respondent. At the household-level, men

and women in the “Other” ethnic category, which includes men and women who

self identify with two or more ethnic categories, commuted 1:49 min more than

white men and women. Men and women whose total income was less than \$25,000

commuted 1:18 min less than men and women whose total income was \$25,000

to \$49,999, while men and women whose total income was \$50,000 to \$74,999,

\$75,000 to \$99,999, and greater than or equal to \$100,000 commuted 1:37, 3:31,

and 3:58 min more, respectively, than men and women whose total income was

\$25,000 to \$49,999. Men and women employed in service and clerical occupations

commuted 1:55 and 1:38 min less, respectively, than men and women employed

Multilevel Models of Commute Times for Men and Women

209

in professional occupations. Men and women who worked part-time commuted

3:55 min less than men and women who worked full-time. Finally, a one unit

increase in the vehicle to worker ratio increased commute times for men and women

by 1:19 min.

Results from the MSA-level of the pooled men–women multilevel model appear

in the Men–Women column of Table 4. Only statistically significant coefficients are

reported. The referent categories for the discrete independent variables population

size, region, and commuter rail represent the typical MSA. At the MSA-level, a one

unit increase in the value of the congestion measure and land area is associated with

an increase in the commute times of men and women of 5:08 and 0:01 min, respectively. Commute times for men and women were, on average, 2:53 min shorter in

large population-sized MSAs than in very large population-sized MSAs. Finally,

a one unit increase in residential density and land use mix score is associated

with a decrease in the commute times of men and women of 0:01 and 0:01 min,

respectively. Overall, results from the pooled men–women multilevel model suggest

that:

men and women who self identify with two or more ethnic categories commute

more than white men and women;

high-income men and women commute more than low-income men and women;

men and women with service jobs commute less than men and women with

professional jobs;

men and women who work part-time commute less than men and women who

work full-time; and

men and women with more access to private vehicles commute slightly more

On average, commute times for the typical male and female in the pooled men–

women subsample were shorter in less congested, large population-sized MSAs.

5.5 Analysis of MSA-Level Residuals from Multilevel Models

Analysis of the MSA-level residuals from the men-only, women-only, and pooled

men–women multilevel models offers valuable information on the geographic variation in the commuting time gender gap. On the one hand, in four MSAs – Austin,

Buffalo, Minneapolis, and San Francisco – commute times for men were longer

than predicted based on the men-only multilevel model and commute times for

women were shorter than predicted based on the women-only multilevel model

(Fig. 2). These four men-longer commute and women-shorter commute MSAs are

located in all four census regions. In seven MSAs – Atlanta, Kansas City, Miami,

New Orleans, Oklahoma City, Philadelphia, and San Antonio – commute times

for men fell into the predicted category based on the men-only multilevel model

and commute times for women were shorter than predicted based on the womenonly multilevel model. These seven men-moderate commute and women-shorter

210

E.J. Zolnik

commute MSAs are mostly located in the South. On the other hand, in one MSA

– Los Angeles – commute times for men were shorter than predicted based on

the men-only multilevel model and commute times for women were longer than

predicted based on the women-only multilevel model. In two MSAs – Pittsburgh

and Portland – commute times for men were shorter than predicted based on the

men-only multilevel model and commute times for women fell into the predicted

category based on the men-only multilevel model. Overall, two of the three menshorter commute and women-longer or moderate commute MSAs are located in

the West. Finally, in two MSAs – Saint Louis and Washington – commute times

for both men and women were longer than predicted; in 11 MSAs – Columbus,

Dallas, Hartford, Houston, Indianapolis, Jacksonville, New York, Orlando, Providence, Seattle, and Tampa – commute times for both men and women fell into the

predicted category; and in five MSAs – Denver, Detroit, Grand Rapids, Milwaukee,

and Phoenix – commute times for both men and women were shorter than predicted

based on the men-only and women-only multilevel models, respectively. Analyzing

MSA-specific residuals extends previous research on geographic variation in the

commuting time gender gap (Wyly 1998) by showing that women’s commute times

are longer than predicted and men’s commute times are shorter than predicted in the

West; particularly in Los Angeles. Analysis of the MSA-level residuals also shows

that men’s commute times were longer than predicted and women’s commute times

were shorter than predicted in San Francisco, which runs counter to Gossen and

Purvis’ (2005) finding of an attenuation in the commuting time gender gap.

5.6 Proportion of Variance Between and Within MSAs

The intraclass correlation coefficient (ICC) is used here to measure the proportion

of variance in private-vehicle commute times between MSAs (Raudenbush and

Bryk 2002; Snijders and Bosker 1999). The ICC is applicable only to randomintercept models such as the men-only, women-only, and pooled men–women

multilevel models reported in this chapter. To estimate the ICC, estimates of

between-MSA and within-MSA variability are substituted for the parameters in the

following equation:

00

D

(3)

2

00 C

where:

is the ICC;

captures between-MSA variability; and

captures within-MSA variability.

00

2

Estimation of the ICCs for the men-only, women-only, and pooled men–women

multilevel models reveal that 0.03%, 0.07%, and 0.05%, respectively, of the variance in private-vehicle commute times was between MSAs. Thus, just as Schwanen

Multilevel Models of Commute Times for Men and Women

211

Fig. 2 Regional differences in commute times from men-only, women-only, and pooled men–

women multilevel models

212

E.J. Zolnik

et al. (2004) found, place/spatial characteristics account for a small proportion of

the total variation in private-vehicle commuting outcomes.

6 Discussion

Comparison of results at the household-level between the men-only, women-only,

and pooled men–women multilevel models tend to support the contention of economic theory (White 1977; 1986) that commute time differences are at least partially

attributable to income and occupational differences between men and women. Total

incomes were, on average, lower in the women subsample than in the men subsample, and men and women in low-income categories commuted less than men

and women in middle- and high-income categories. However, commute times for

women in the highest-income categories were longer by 3:07 and 2:09 min, respectively, in comparison to commute times for women in the middle-income category.

This finding suggests that higher incomes were more synonymous with longer commutes for women than for men. Occupation had different effects on commute times

for men and women. None of the occupational categories had an effect on commute

times for men. But, as expected, commute times for women employed in femaledominated industries such as service and clerical were shorter than for women

employed in professional occupations. Such a result is consistent with other studies

where shorter commute times for women employed in female-dominated industries have been reported (Wyly 1998). Interestingly, commute times for women

employed in male-dominated, manufacturing occupations were longer than for

women employed in professional occupations. Such a result is not consistent with

results from a study conducted in Philadelphia, for example, where commute times

for women were found to be shorter than commute times for men, regardless of the

private vehicles contributes to private-vehicle commute time differences between

men and women. Access to private vehicles was lower for women in the womenonly subsample than for men in the men-only subsample. However, the association

between private-vehicle commute times and the ratio of vehicles to workers was

positive for both men and women, and greater access to private vehicles appeared to

lengthen commute times for men more than for women.

Comparison of household-level findings across all models offers no support for

the household responsibility hypothesis (Turner and Niemeier 1997); none of the

lifecycle-stage categories for women were statistically significant in the womenonly multilevel model. Interestingly, even though the percentage of women in the

women-only subsample who worked part-time was higher than the percentage of

men in the men-only subsample who worked part-time, private-vehicle commute

times were shorter for men (3:35 min) and women (3:01 min) who worked part-time

by approximately the same amount. Such a result suggests that part-time work had

the same effect on private-vehicle commute times for men and women.

Multilevel Models of Commute Times for Men and Women

213

Comparing model results at the MSA-level indicates that congestion had the

largest differential effect on the commute times of men and women. Congestion,

measured using the TTI, is associated with an increase of 7:07 min in private-vehicle

commute times for men in the men-only subsample and a decrease of 11:11 min for

women in the women-only subsample. In contrast to these results, commute times

for women were, on average, shorter in Northeastern and Midwestern MSAs than

in Western MSAs where congestion was highest. Taken together, the large differential effect of congestion on commute times for men and women appears to be

a phenomenon specific to Los Angeles, where congestion was ranked first among

the 43 MSAs, and where commute times were shorter than expected for men and

longer than expected for women based on the men-only and women-only multilevel

models, respectively. Land area was associated with an increase in commute times

for men and women by the same amount. As expected, sprawl, measured here as a

function of residential density and land use mix, had negative effects on commute

times for men and women. That is, higher residential density and better land use

mix appears to lower commute times.

The coefficient estimates for land area, residential density, and land use mix are

statistically significant, but the strength of the associations between each of these

variables and private-vehicle commute times, for men and women, is very small.

Interestingly, the degree of centering, which predominantly reflects job sprawl,

was not statistically significant. Such a result tends to contradict arguments that

job sprawl in urban labor markets contributes to the commuting time gender gap

(Wyly 1998). Finally, commute times for women in MSAs that did not have commuter rail were shorter than commute times for women in MSAs that did have

commuter rail. Such a result suggests that the absence of public alternatives leads

to more private-vehicle commuting which is more time efficient for longer distance

commutes.

7 Conclusions

A reassessment of home-work linkages by Hanson and Pratt (1988a) underscores

the need to consider how the, “home-work link functions for a variety of diverse

household types in a variety of local contexts” and “to make explanations scale

specific” (p. 318). In total, the results reported in this chapter suggest that compositional effects such as income and occupation have a greater impact on variation in

commute times for men and women than contextual spatial effects such as sprawl.

As such, results from the chapter are more supportive of economic rather than

household responsibility explanations for the commuting time gender gap. Further,

the results point away from polycentricity and job sprawl as major contributors to

the commuting time gender gap (Rosenbloom 2006). Nonetheless, one contextual

effect – congestion – has a large differential impact on commute times for men

and women especially in Los Angeles. Likewise, regional variations in commute

times for women are evident – commute times are shorter for women in the South

### Tài liệu bạn tìm kiếm đã sẵn sàng tải về

1 Descriptive Statistics for the Household- and MSA-Level Dependent and Independent Variables

Tải bản đầy đủ ngay(0 tr)

×