Results



 ^K

k

k k

w w t

e

1 (9) where K denotes the number of street types; w denotes the metabolic rate while walking _k on a type k street, and t denotes the walking time on a type k street. _k

3.3 Results

Then a regression analysis was performed using SPSS 12.0 to establish the predictive model

 

 . The summary of the values for the observed terrain factors for the seven street sites are shown in Table 3.4. The regression analysis included two steps: the first step was to examine the statistical insignificance among the candidate variables. After deleting the insignificant variables, the second step in designing the predictive model was performed. As shown in Table 5, there is a difference in the ranking of the mean observed values between daytime and evening. In addition, each site shows substantial variation, implying that individual characteristics and timing may contribute at least in part to the effects. Thus, in addition to the five street factors, gender, VO_2max, timing and temperature are also involved in step 1.

It must be noted that the coding method determined the final number of variables used in regression analysis. For example, we treated land use as a categorical (nominal) variable and coded it as a dummy variable, and it is therefore examined for two types of land use.

Planting is assumed to be an ordinal variable because there is wide agreement that a greener area is more pleasing to pedestrians. A more detailed explanation of those variables is discussed in Table 6. A total of 10 candidate variables are used in the first step of the regression analysis.

Other coding methods such as effect coding are worth considering, but they must be designed to fit the surveyed pedestrian environment. For example, when using effect coding, the question is usually termed as “good, fair, poor”, and the codes may then be set as “+1, 0, -1”. However, this is more suitable for psychological measurement. Regarding planting (PL) in this research, green areas are available in most of our surveyed streets, so the question design actually focuses on how green the street is. Coding potted plant as -1 seems to be

24

unsuitable.

Table 6 shows that two personal physiological factors, gender (GE) and VO_2max, are statistically significant. This suggests that the value of the terrain factor will significantly vary with gender and VO_2max. A negative sign for the gender variable indicates that, when all other factors are equal, women expend more energy than men. It is probable that women generally have smaller muscle mass and produce less power, thus they put greater effort into the same movement. Kerr et al. (2001) found a similar phenomenon in stair-climbing: after placing motivating messages on the stair risers, men are significantly more willing to use stairs than women. Their finding confirms the effect of gender on WTW. VO_2max proves to be strongly significant, suggesting that physical fitness is a determining factor for energy expenditure, and that VO_2max reflects individual physical fitness rather than age.

Right of way (ROW) was found to be statistically significant, but the width of the walking space was not. These findings suggest that the range of walking space width (WD) from 1.5 to 5 m does not significantly affect the terrain factor. This is probably due to the fact that the testing range of the width of the walking spaces being tested was wide enough for most participants to walk unhindered. On the other hand, the disturbance from vehicles and incompatible activities significantly affected a pedestrian’s energy expenditure. Lighting (LG) had a negative sign and was significant, suggesting that good street lighting contributes greatly to the street amenity. It also suggested that heart rate increases significantly at night but declines when the lighting level is higher. Planting (PL) also shows a significant effect on WEE, confirming that the larger the green area, the higher the amenity of the street. Temperature is not statistically significant, possibly because the participants were used to the range of 18~28℃ weather and temperature did not significantly affect energy expenditure.

Walking in the evening (NI) causes a greater terrain factor value than during the daytime. It is possible that individuals must make more effort to understand the environment or are more concerned about personal safety when lighting levels are low. The greater effort results in a higher heart rate. Both land use factors (CM and MU) were not statistically significant, indicating that land use patterns did not lead to a significant difference in energy expenditure between the seven streets in our experiment.

25

At the second stage, we removed the insignificant variables and developed a terrain factor predictive function. This function was designed because some variables had independent effects while others reflected interactive effects. For example, lighting is used mainly after dark, and therefore the best calibrated predictive function of an adjusted terrain factor is

where the fourth and fifth terms are designed for variables with interactive effects in which NI is a dummy variable to reflect walking after dark (NI=1) or during daytime (NI=0). The multiplication term of lighting-evening means that street lighting reduces friction after dark by enhancing visibility, but it only operates when walking during the evening. The variance inflation factor (VIF) is a statistical index that measures how much the variance of an estimated regression coefficient is increased because of collinearity. The values in Table 6 show that the variables in the predictive function do not have significant collinearity (VIF＜

10). Since NI and LG are designed as interactions, compared to a linear function, the VIF for NI increases from 1.026 to 5.639 and for NI．LG it increases from 1.47 to 5.910, respectively.

Table 5 Experiment outcome summary.

Day Evening

Experiment site nˆ tau c nˆ tau c

1(Arterial road) 0.798±0.393 0.230(0.076) 1.130±0.631 0.314(0.034) 2(Boulevard) 0.661±0.342 0.305(0.047) 0.833±0.476 0.345(0.052) 3(Residential street) 0.607±0.311 0.208(0.069) 0.841±0.474 0.510(0.001) 4(Commercial parkway) 0.592±0.283 0.317(0.035) 0.722±0.395 0.238(0.062) 5(Mixed-use street) 0.755±0.367 0.289(0.054) 0.901±0.485 0.328(0.002) 6(Downtown street) 0.738±0.363 0.078(0.605)* 0.938±0.466 0.125(0.500)*

7(Alley) 0.696±0.361 0.355(0.008) 1.067±0.607 0.322(0.026) Note: The adjusted terrain factor values (nˆ) are shown asmeanSD; tau c is shown as value (p-value) ; * not significant at the 10% level.

To determine the consistency between psychological perception and physiological response, the participants ranked all streets by preference (1~7). We then compared the data of the observed ranking of the terrain factor by means of Kendall's tau c test. As shown in Table 5,

26

most sites show a statistical consistency between psychological and physiological rankings at the 10% significance level, but site 6 does not. This discrepancy may be due to the unexpected effects site 6 had on some of the participants. Site 6 is a downtown street with a variety of activities and actors in a compact arcade, such as street vendors. We may have overlooked the fact that some positive emotions such as excitement also cause a high heart rate. However, this is beyond the scope of this research, and we will leave that to future study.

Table 6 Results of regression analysis.

Step 1 Step 2

Coefficients Significance Coefficients Significance VIF

GE -0.537 0.012 -0.528 0.005 1.178

2max

O

V 0.037 0.02 0.036 0.001 1.181

ROW 0.12 0.016 0.152 0.011 1.194

WD(m) -0.03* 0.131 － － －

PL -0.058 0.032 -0.074 0.006 1.09

LG -0.029 0.048 － － －

CM 0.095* 0.202 － － －

MU 0.029* 0.584 － － －

NI 0.222 0.004 0.408 0.001 5.639(1.026)

Temperature -0.006* 0.314 － － －

Constant 0.031* 0.877 -0.386 0.032 －

NI×LG － － -0.02 0.015 5.910(1.47)

Adjusted R² 0.722 0.721

Note: Step 1 is performed to examine statistical significance of the variables, * removed when not significant at the 5% level; step 2 is performed to design the regression model.