The Impact of Weather Conditions on Mode Choice: Empirical Evidence for the Netherlands

The Impact of Weather Conditions on Mode Choice: Empirical Evidence for the Netherlands Muhammad Sabir*, Mark J. Koetse and Piet Rietveld Department o...
Author: Barrie Parks
3 downloads 1 Views 184KB Size
The Impact of Weather Conditions on Mode Choice: Empirical Evidence for the Netherlands Muhammad Sabir*, Mark J. Koetse and Piet Rietveld Department of Spatial Economics Vrije Universiteit Amsterdam * Corresponding author: VU University, Department of Spatial Economics, De Boelelaan 1105, 1081 HV, Amsterdam, The Netherlands, E-mail: [email protected].

Abstract This paper investigates the influence of weather conditions on individuals’ mode choice decisions. The major contribution of this study is that it uses micro level data from 1996 on transport behavior and weather conditions covering the entire Netherlands. Moreover, weather conditions are measured on an hourly basis. We use a multinomial logit model which allows for an accurate investigation of the effect of weather conditions on mode choice decisions. In (extremely) low temperatures, people switch from biking to car and public transport, whereas people prefer walking and biking as temperatures increase. Strong wind negatively influence the use of the bicycle, while precipitation also causes a reduction of the biking trips and raise in the car trips.

1.

Introduction

The choice of transport mode is an important issue in transport planning. Mode choice affects the general efficiency with which one can travel and the amount of space devoted to transport functions (Mandel et al. 1997). Mode choice decisions are influenced by many factors, such as travel time, travel costs and socio−economic characteristics of the trip maker. This decision is also sensitive to the occurrence of unforeseen and key events such as accidents, extreme weather, and substantial changes in one’s personal life, like a change of residence, a change of workplace and children (Van Waerden et al. 2002). All these factors are expected to motivate changes in mode choice decisions. The role that weather plays in mode choice decisions may change due to changes in the climate system.1 Global temperature is predicted to rise by 2o C to 3o C within the next fifty years, on the basis of current temperature trends, and furthermore by 5o C by 2100 (Stern 2006). Oldenborgh and Ulden (2003) find that temperature in De Bilt in the Netherlands has risen by 1K over the 20th century, which is quite close to global

1 Climate changes are due to human and non−human reasons. The Difference between weather and climate is that weather is a combination of events in atmosphere and climate is overall accumulated weather in a certain location.

2 temperature increase, as shown in Figure 1.2 Annual precipitation in the Netherlands has increased by 20% since 1900 (KNMI 2006). The frequency of extreme weather will increase due to global warming (Stern 2006).

Figure 1. Comparison of global temperature and temperature in The Netherlands (K) form 1900 to 2000 KNMI climate changes scenarios for the Netherlands suggest that temperature in the Netherlands continue to rise over the next 50 years (KNMI 2006), i.e., milder winters and warmer summers. Moreover, winters will become wetter and extreme precipitation will increase. Extreme showers during summer will also become more common, although the number of rainy days in summer will decrease. Calculated changes in wind speed are expected to be small as compared to natural fluctuations. These changes likely have serious implications for mode choice decisions. The three prominent modes of transportation in the Netherlands are car, bicycle and walking. These three modes cover more than 90% (CBS) of all Dutch individual trips. Walking and biking are much more vulnerable to weather conditions, especially in extreme weather situations. Since global temperature will increase during the coming decades and extreme weather events will become more frequent, weather is likely going to play a more dominant role in transportation in general and in mode choice decisions in particular. Many studies are available on the relationship between weather and transport. Most of these studies focus on road safety (e.g., Eisenberg 2004; Andrey et al. 2001; Edwards 1996) or vulnerability of transport infrastructure to extreme weather conditions (e.g., Waalkes 2003). Stern and Zehavi (1990) use a seven years dataset from Arava road in Israel and found that road accidents are more frequent during medium and high heat stress periods as compared to normal periods. Also, fog and wind may have an increasing effect on the number of accidents (see, e.g., Hermans et al. 2006; Edwards 1996). Suarez et al. (2005) study potential impacts of climate change on the performance of surface transport in the Boston Metro Area with a focus on transportation disruption. They use projected 2

The station De Bilt is representative for the mean climate conditions in the Netherlands. It temperature and wind direction records is considered the most homogeneous long−term records of the Netherlands. (Jan Van Oldenborgh and Add Van Ulden 2003)

2

3 changes in land use and changes in demographic and climate conditions in urban transportation modelling, and found that global warming doubles delays and loss of trips due to additional riverine and coastal flooding. Winters et al. (2007) investigate the impact of climate and demographic characteristics on utilitarian cycling in Canadian cities for year 2003. There findings shows that utilitarian biking are negatively influenced by increasing numbers of days with precipitation and freezing temperatures. Furthermore, older age, female gender, lower education and higher income are associated with lower likelihood of cycling. The literature that focuses specifically on the impact of weather on mode choice decisions is scarce. Khattak and De Palma (1996) studied traveller behaviour in Brussels under normal and unexpected travel conditions. Using data on individual characteristics and weather conditions for 1992, they find that adverse weather causes changes in mode choice, route choice and departure time of automobile commuters. They furthermore found changes in departure time due to adverse weather conditions to be of more importance for automobile commuters than changes in route and mode choice. Automobile users constrained by family commitments and those who drove alone were found to change their mode choices less often due to bad weather. De Palma and Rochat (1999) conduct a similar survey among Geneva commuters. The patterns found are similar to the ones found in Khattak and De Palma (1996). Adverse weather leads to changes in mode choice, route choice and departure time choice, with the latter again being most important. A more recent study by Aaheim and Hauge (2005) uses micro−level information on individual transport behaviour in Bergen (Norway). First, they study the impact of weather conditions on mode choice decisions at an individual level using a quantal response model. They extend the analysis to the regional and national level by using simulated weather forecasts for Norway. The main finding was that increases in precipitation and wind increase the likelihood of use of public transportation as compared to walking and biking. For the regional level, their analysis shows that weather conditions do not induce a switch between public and private transport. Furthermore, at the macro level the impact of climate change on travel patterns appears small for Norway. Despite their useful insights, these studies have some major limitations. First, the relatively short survey periods in Aaheim and Hauge (2005), De Palma and Rochat (1999) and Khattak and De Palma (1996) imply that there is relatively little variation in weather conditions. It is therefore difficult to draw general conclusions from these studies. More specifically, if the focus of research is on the general impact of climate change on transport choices, periods covering only a few months are insufficient, since climate changes likely have a differential effect on weather conditions in different seasons. Second, the weather indicators used in these studies were recorded once a day, or only a few values of a limited number of weather indicators were available. In countries in which weather is subject to hourly changes, such as the Netherlands, such an approach is not viable. Furthermore, the numbers of observations used in these studies are small. This makes it even more difficult to generalise the findings. Clearly, there is a need to analyse the influence of weather conditions on mode choice decisions using data that have a large coverage in terms of geographical location, time duration and more suitable measure of weather indicators. Therefore, the aim of this paper is to study the influence of weather conditions on individual travel behavior, while using data that answer to all of these criteria. An important contribution of this paper is that 3

4 hourly data are used for all weather variables and that the data cover the entire Netherlands for 1996. The remainder of the paper is organised as follows. Section 2 contains explanation of data and its sources. This section also provides a detailed discussion of the explanatory variables and explains the procedure use to interlink independent data sets used in the analysis. Furthermore, this section presents some descriptives statistics from the data and some initial idea of effects of weather on mode choice of individuals through graphs. In addition, we discuss the model used. Section 3 presents the model results and a discussion. Finally, Section 4 presents conclusions and topics for further research.

2.

Data and model specification

The data used in this paper were taken from several sources. First, we make use of a transport survey of Statistics Netherlands for 1996. Over the course of an entire year, more than 600,000 individuals from different municipalities in The Netherlands were asked to fill out a questionnaire on their travel behaviour during a certain day and important household and individual characteristics. The questionnaire contained seventy−seven different questions about trips made during a single day and on important individual and household characteristics. We could not make use of all observations because of missing values. After removing the observation with missing values, the sample consists of around 530,000 observations. Table 1 shows the percentage share of each mode in data. Table 1: Percentage of Mode share Mode Percentage Share Walk Bicycle

20.40 26.83

Car Public transport Other Total

45.91 4.34 2.52 100

The second data source is a weather database of the Royal Netherlands Meteorological Institute (KNMI) for 1996. The weather database contains hourly weather data for every day in 1996. The data were recorded by 37 weather stations, more or less evenly spread throughout The Netherlands, and cover all 458 municipalities of the Netherlands. The weather data include variables like hourly temperature, average wind speed, maximum wind speed, amount of precipitation, duration of sunlight, etc. Table 2 contains information on monthly weather indicators for 1996. The data sets are independent and were linked in such a way that each trip observation was assigned the weather conditions of the hour in which the trip took place, and from the weather station that was nearest to the municipality of departure. The underlying assumption is that the trip maker base his or her decision on the weather conditions that prevailed at the moment (hour) of departure. 4

−6.2

−8.1

−0.7

4.9

6.1

6.9

3.9

0.2

−4.1

−14.6

March

April

May

June

July

August

September

October

November

December

9.6

15.5

22.1

21.8

31.6

30.4

34.3

27.6

27.3

13.7

10.9

0.84

6.30

11.98

14.25

19.19

17.65

17.46

11.94

11.59

4.28

0.58

0.3

5.7

12.2

14.4

18.9

17.2

16.4

11.4

11.4

4.1

0.5

4.86

4.00

2.62

2.77

3.71

3.43

5.07

3.83

5.70

3.29

4.10

170

190

190

170

160

190

139

144

149

144

195

43.86

50.12

42.13

43.18

40.51

46.12

42.26

48.18

42.39

47.56

51.30

26.26

31.40

26.99

24.38

21.42

22.44

20.06

21.65

21.23

22.37

29.07

3.6

13.1

8.30

6.1

21.0

22.3

53.7

10.7

4.5

3.2

4.8

0.06

0.20

0.06

0.05

0.17

0.07

0.05

0.06

0.01

0.03

0.07

0.28

0.58

0.30

0.26

0.82

0.49

0.77

0.35

0.08

0.14

0.31

Precipitation (mm/hour) Max Mean S.E 1.5 0.01 0.06

Note: The minimum wind speed and minimum precipitation are not shown in the table as they are zero for all months

−20.6

February

Table 2: Monthly weather indicators for 1996 in The Netherlands Temperature (C) Wind speed (m/s) Months Min Max Mean Median S.E Max Mean S.E January −12.1 12.3 0.21 −0.5 4.86 175 54.62 24.45

5

5

6

Figure 2:

Temperature and Mode Choice

Figure 3: Wind Strength and Mode Choice

Figure 4: Precipitation per hour and Mode choice

Figure 5: Mode choice during various parts of the day

6

7

Bar charts of mode choice and weather variables were analyzed to give a preliminary ideas of a relationship between weather and mode choice decision. The percentage share of trips of each mode in total trips for different weather indicators are shown in Figures 4, 5 and 6. An increasing percentage share of bicycle trips can be observed directly with increasing temperature; however, this increase is small in last category. On the other side, the percentage share of trips made by car is less in high temperature. The percentage share of other modes although changes with changing temperature, a clear pattern is difficult to observe from Figure 2. Figure 3 shows that percentage trips made by bicycle decline marginally with increase in wind strength. The share of car trips fractionally increases with increase in wind strength. Furthermore, the percentage share of trips made by public transport, walking and other category varies with increase in wind strength although the pattern is not much clear from Figure 3. Finally, Figure 4 shows a comparatively decreasing percentage share of bicycle trips in increasing rain whereas for car trips the reverse happens. The percentage share of walks appears to be not affected. Moreover, minor changes can be noted in the percentage share of trips made by public transports and other category. These bar charts give motivation for analyzing the data further by an econometric model. With respect to the model specification, let’s assume that individual i select choice j for transportation. The probability for this to happen is equal to the probability that the utility level associated with choice j exceeds the utility level associated with all other available alternatives. In other words, by selecting mode j, the decision maker maximizes his utility, i.e.:

(

)

Prob (Yi = j) = Prob U ij > U ik for k ≠ j .

(1)

The utility of the trip taker is now given by: U ij = Vij + εij ,

(2)

where Vij represents the deterministic part of the utility and ε ij represents the stochastic part. The deterministic part is a function of individual characteristics and weather conditions with unknown parameters βi . By substituting equation (2) in equation (1), we get:

(

)

(

)

Prob (Yi = j) = Prob Vij + εij > Vik + εik = Prob εij − εik > Vij − Vik for k ≠ j .

(3)

If the disturbance terms are independent and identically distributed, then the error terms follow the Weibull distribution (McFadden 1974), so the probability of selecting mode j is given by:3

3

The Weibull distribution has the property that the cumulative density of the difference between any two random variables with this distribution is given by the logistic function.

7

8

Prob (Y = j ) =

exp (V j ) J

∑ exp (V ) k =1

.

(4)

k

Equation (4) represents what is called a Multinomial Logit Model. Now, as Vij can be written as a function of explanatory variables x with unknown parameters β , the following specification is used for the estimation of the utility function: N

V j = α j + ∑ βn, j xn + ε j , n =1

(5)

where α j is a transport mode specific constant, j represents transport mode, and N is the number of explanatory variables. Note that this implies that the coefficients on the explanatory variables vary across modes. In our model we assume that each individual faces the same choice set for transportation, which consists of five different modes, i.e.:

    

Walking (reference category); Bicycle; Car; Public transport (bus, tram, subways, train); Other (moped, motor, scooter, taxi, truck, delivery van).

Details of the explanatory variables used in our model are given in Table 3. Individual characteristics are controlled for by including the variables age, income, gender and car ownership. The variable urbanization was included as a continuous variable which has five possible values, ranging from 1 for most urbanized areas to 5 for rural areas. We also included distance of the trip, for obvious reasons. Furthermore, dummies were included to control for seasonal variations and to control for working and non working days of the week. In the original dataset, eight different trip purposes are distinguished. We reduced this set to four major categories based on some common characteristics. The first category consists solely of commuting trips, and is included as reference category. The second category consists of shopping trips, trips to visit family and friends, recreational trips and sports trips. These are combined and named as recreational and sports trips. The business and educational trips are the other two categories. These two categories are placed in separate categories because of their difference with other categories. Therefore, we end up with four major trip purposes. Differences in mode choice for different trip purposes are controlled for by including dummy variables for categories two, three and four.

8

9

Table 3: Details of explanatory variables Variable Distance (D)

Details and Measurement Distance of trip measured in kilometers. It is the distance covered in a single trip made by individuals.

Income (I)

Log income of households measured in thousands of Guilders.

Age (A)

Age of the trip taker (years). Age is expected to have non−linear effects so it is used as non−continuous variable. Three categories of age are distinguished. one for people up to 18 years old, people between 18 and 60 years and people older than 60 years.

Gender (G)

Gender (Dummy = 1 for male).

Car Ownership

Car ownership (Dummy =1 for having car)

Urbanization (U)

Degree of urbanization. The categories of urbanization are; very urban, urban, moderately urban, little urban, rural.

Trip Purpose ( BS, ED and RS)

Trip purposes are dummy variables. We create a dummy for commuting trips (reference category), a dummy for business trips (BS), a dummy for educational trips (ED), and a dummy for recreational and sports trips (RS).

Hour of departure (H)

Time of day/night during which trip took place. It is also used as non−continuous variable in order to control for peak hours and day and night effects (Dummies). Four dummies are used for different hour categories.4

Temperature (T)

Temperature during hour of departure (o C). Six dummies were used representing different categories of temperature ranging from below − 8o C to 25 o C and higher.5

Wind (W)

Wind is measured in Beaufort, also known as wind strength.

Precipitation (P)

Precipitation during hour of departure (0.1 mm per hour).

Week day

Two dummies are used for days during the week. One dummy for weekends (reference category), one dummy was used for working days of week.

Seasonal variable

Four dummies were used for controlling seasonal variations. These were a dummy for spring (if month is March, April or May), a dummy for Summer (if month is June, July or August), a dummy for Autumn (if month is September, October or November) and a dummy for winter (if month is December, January or February). Dummy for Spring is used as a reference category.

4 The categories for hours are between 07:00 and 10:00 (reference category), between 10:00 and 17:00, between 17:00 and 19:00, and between 19:00 and 07:00, respectively. 5 The categories for temperature are temperature less than –8o C, temperature between –8o C to 0o C , temperature between 0o C to 10o C, temperature between 10o C to 20o C, temperature between 20o C to 25o C and temperature greater than 25o C, respectively.

9

10

We included the hour of departure in the model in order to control for differential effects of travel during day and night and travel during for peak and non−peak hours. Four dummies are used for the hour of departure, which include dummies for the peak hours (morning peak from 07:00 to 10:00 and evening peak from 17:00 to 19:00) and two other dummies one covering the remaining part of the day and other covering nighttime. We take account of the possibility that the effects of weather conditions on mode choice decisions are non−linear, so for most weather indicators we distinguish between several categories.6 Three precipitation categories were distinguished, i.e., no precipitation (reference category), precipitation upto 1 mm, and precipitation greater than 1 mm. Wind strength is measured in Beaufort scale. 7 Including a continuous variable as wind speed in the model did not reveal a clear pattern. In order to assess whether high wind speed have an effects, we distinguish two categories for wind strength namely, wind strength up to 6 Bft and wind strength equal to or higher than six Bft. The effects of temperature were analyzed by making 6 temperature categories i.e., less than − 8o C, between − 8o C to 0o C, between 0o C and 10o C (reference category), between 10o to 20o C, between 20o to 25o C, and greater than 25o C. It is expected that the effects of temperature are non-linear especially on both extremes. Furthermore, this division of temperature categories will result in a clearer picture of the individual travel pattern in different temperature and weather conditions. Details of the explanatory variables used in the model are given in Table 3.

3.

Estimation results

We estimated five different models. One model was estimated for all trips while controlling for all explanatory variables shown in Table 2. We also estimated the model for four different categories of trip purposes. Since the coefficients of a Multinomial Logit model are not interpretable directly we report the marginal effects of the full model and the associated standard errors in Table 4.8 The marginal effects of the other four models are given in Appendix B. Coefficients in bold imply statistical significance at the 5% level of significance. A positive sign of a marginal effect represents an increase in the probability of selecting that particular mode. The marginal effects of distance, income, gender, urbanization and car ownership are significant for all modes. The age coefficients show that younger people are forced (or like) to use the bicycle more then the old and middle-aged people. The use of a car is marginally 6

We did not include sunlight as weather indicator as it effects are covered by hour variables. The Beaufort scale measures the wind strength. This measure scales wind speed on a scale of 1 to 12. On this scale, 6 BFT represents powerful wind having speed in the range of 39 to 49 kilometers per hour (or between 10.8 to 13.8 meter per second) over a period of 10 minutes. Similarly, 12 BFT represents a hurricane in which the speed of wind is greater than 117 kilometers per hour (or greater than 32.6 meter per hour). See www.knmi.nl. 8 We also made a distinction between trips from home and trips from other places. The results on the impact of weather conditions were very similar to the results presented above. 7

10

11

Table 4: Marginal effects of the full model (standard errors in parentheses)*

Distance (C) Business (D) Educational (D) Recreational and Sports (D) Income (C) Urbanization (D) Age02 (D) Age03 (D) Gender (D) Car ownership (D) Hour † Hour(10:00 to 17:00) (D) Hour(17:00 to 19:00) (D) Hour(19:00 to 07:00) (D)

Working Day (D) Summer (D) Autumn (D) Winter (D) Wind † Wind greater than 6 BFT (D) Temperature† Less than −8o C (D) Between −8o to 0o C (D) Between 10o to 20o C (D) Between 20o to 25oC (D) Greater than 25o C (D)

Walk

Bicycle

Car

.2676 (.0013) −.0061 (.0049) −.0006 (.0016) −.0031 (.0026) .0075 (.0004) .0051 (.0006) .1455 (.0019) .1424 (.0030) −.0290 (.0015) .3477 (.0034)

Bus, Tram, Subways, Train .0222 (.0001) −.0002 (.0016) −.0005 (.0005) −.0009 (.0009) −.0016 (.0001) −.0041 (.0002) .0052 (.0007) .0028 (.0010) .0043 (.0005) −.0488 (.0007)

−.2311 (.0005) .0051 (.0022) .0006 (.0007) .0004 (.0012) −.0040 (.0002) −.0096 (.0002) −.0262 (.0009) .0149 (.0013) .0072 (.0007) −.0900 (.0014)

−.0740 (.0012) .0029 (.0040) −.0011 (.0013) .0028 (.0022) −.0030 (.0003) −.0084 (.0005) −.0965 (.0016) −.1266 (.0025) .0270 (.0012) −.1857 (.0025)

.0052 (.0009) −.0211 (.0011) −.0003 (.0013)

Other

.0154 (.0002) −.0017 (.0016) .0015 (.0005) .0008 (.0008) .0011 (.0001) .0001 (.0002) −.0280 (.0005) −.0335 (.0010) −.0095 (.0004) −.0232 (.0007)

−.0315 (.0016) −.0401 (.0019) −.0771 (.0023)

.0513 (.0020) .0847 (.0023) .1123 (.0027)

−.0237 (.0006) −.0209 (.0007) −.0335 (.0009)

−.0013 (.0006) −.0027 (.0007) −.0014 (.0008)

.0238 (.0008) .0034 (.0012) −.0064 (.0010) .0075 (.0012)

−.0260 (.0014) .0045 (.0020) .0033 (.0017) −.0161 (.0021)

.0121 (.0017) −.0063 (.0025) .0002 (.0021) .0070 (.0025)

−.0089 (.0006) −.0022 (.0009) .0038 (.0007) .0048 (.0008)

−.0010 (.0005) .0005 (.0007) −.0009 (.0006) −.0032 (.0008)

.0147 (.0026)

−.0381 (.0049)

.0154 (.0058)

.0065 (.0017)

.0014 (.0018)

.0255 (.0035) .0172 (.0016) −.0061 (.0016) .0055 (.0026) .0072 (.0010)

−.0849 (.0068) −.0558 (.0029) .0265 (.0028) .0504 (.0043) −.0380 (.0017)

.0638 (.0078) .0410 (.0034) −.0184 (.0035) −.0530 (.0057) .0321 (.0021)

.0044 (.0023) .0039 (.0011) −.0031 (.0013) −.0042 (.0022) .0025 (.0007)

−.0088 (.0029) −.0063 (.0011) .0012 (.0010) .0013 (.0016) −.0038 (.0006)

11

12

Precipitation† Less then or equal to 1 mm (D) Greater than 1mm (D)

.0029 −.0302 .0250 .0019 .0033 (.0013) (.0025) (.0029) (.0009) (.0009) −.0009 .0033 −.0398 .0432 −.0058 (.0026) (.0048) (.0056) (.0018) (.0020) † Reference category for hour, temperature, wind and precipitation are; hour between 07:00 to 10:00, temperature between 0o and 10oC, wind up to 6 BFT, and no precipitation, respectively. * Bold coefficients are statistically significant at 5%.

higher for middle aged people than for older people. Older people walk more as compared to the other two age groups. The marginal effects of distance shows that the probability of selecting the car increases by 26.7% and probability of public transport increases by 2.2% respectively with each additional kilometer of distance. The probability of choosing to walk and bicycle decreases by 23.1% and 7.4% respectively, for every additional kilometer of distance. The marginal effects of urbanization reflect the fact that cars are preferred to all other modes in rural areas. However, the effects of the degree of urbanization on mode choices are small. The probability of using the car is high for individuals with high household incomes. The probability of selecting walking and biking reduces with increasing household income. The marginal effects of car ownership show that there is a 34% higher probability of traveling by car for households with a car than without a car.9 The probability of using the car by males is 2.9% less than females. This indicates that females use more cars as compared to males. The marginal effects on the hour of departure variables show that the probabilities of selecting the bicycle and public transport as transport mode are high during morning peak hours as compared to other hours during the day and night. The opposite holds true for car use. The marginal effects of weekday show that people tend to use the bicycle and public transportation more on weekends. The marginal effects of seasonal variation are statistically significant but small. With respect to the impact of weather conditions, the marginal effects of temperature show some interesting patterns. First, compared to normal temperatures (temperature between 0o C and 10o C), the probability of selecting the bicycle decreases by 8.4% when temperatures are very low (temperature below –8o C) and by 5.5% when temperatures are low (temperature between –8o C and 0o C). In other words, this shows that the absolute share of bicycle trips in the total number of trips reduces from 26% to 18.3% and 20.5% for extreme low temperatures and low temperatures, respectively. This is finding is consistent with our expectation, since it is quite hard to do biking in low temperatures as compared to normal temperatures. In contrast, compared to normal temperatures, the probability of using the car increases by 6.3% and 4.1% when temperatures are very low and low, respectively. This indicates that the share of car trips in total trips increases from 45.9 % to 52.2% and 50% for extreme low and low temperatures respectively. In other words, in low temperatures about half of individuals’ trips are made by car. The results also show an increase in the probability 9

In order to control for the fact that people and households may not have access to a car we opted to go with the simple solution of including a car ownership dummy in the model specification. Probably a better way to control for this is by reducing the choice set for those households.

12

13

of walking of around 2.5% when temperatures are below 0o C. These patterns clearly show a decrease in bicycle use and an increase in car use and walking when temperatures fall below 0o C. These results are consistent with the findings of Winters et al. (2007) which concluded that utilitarian biking are influence negatively by increasing numbers of days with precipitation and freezing temperatures. Our present analysis does not give any information about the changes in total travel demand so we cannot ensure whether these patterns reflect the shifting from one mode to another or changes in overall travel demand. Interesting is that when temperatures are higher than normal, the pattern is reversed, i.e., in temperatures higher than 10o C the use of car is decreases whereas the use of the bicycle increases. Changes in the use of public transportation and walking are small. A final interesting result is the effect on mode choice for temperatures above 25o C. The marginal effects show a decrease in probability of biking and an increase in probability of the car and public transport. However, it is not clear that this pattern is result of changes in overall demand or due to substitution among different mode choices. Marginal effects of wind are statistically significant for all modes expect for the ‘other’ category. The probability of bicycle use decreases by 3.8% when wind strength is above 6 BFT, compared to wind strengths below 6 BFT. This indicates that the share of biking in total individual trips decreases from 26.8% to about 23%. On the other hand, the probability of using the car and public transport increases by 1.5% and 0.65%, respectively. This shows that high wind strengths discourage the use of the bicycle. The probability of selecting the bicycle decreases by 3.0% if precipitation during the hour of departure is less than 1mm, and by 3.9% if precipitation is more than 1mm, compared to a situation in which there is no precipitation. The probability of selecting the car increases by 2.5% and 4.3% respectively, for the same amounts of precipitations. This implicitly shows that the share of the car in total number of trips increases from 45.9% to 47.4% and 52.2% for precipitation less than 1mm and greater than 1mm, respectively, as compared to no precipitation. This pattern reveals the tendency of decrease in the bicycle use and an increase in the car use for increasing amounts of precipitation. Finally, marginal effects for public transportation are statistically insignificant, while the impact of precipitation on the propensity to walk is small. The models estimated for different trip purposes sketch a similar picture of the effects of weather on individual mode choice decisions; although the magnitude in the percentage shares effects are different for different trip purposes. Wind effects are significant for bicycle and car use in the models estimated for commuting, educational and recreational, and sports trips, which shows a decreasing probability of bicycle use for wind strengths higher than 6 Bft. In business trips, winds effects are significant for only walking and bicycle use. Biking is discouraged by high strengths wind for business trips, whereas the probability of walking increases. It may be noted, however, that the reduction in probability of selecting the bicycle is higher for business trip as compared to other trips purposes. Similar to wind strength, the influence of temperature shows a similar general pattern for all trip purposes, although the effects are different in magnitude. For instance, the probability of selecting the bicycle for business reduces by 15% and 8.4% for extremely low and low temperatures respectively, as compared to normal temperature. This reflects that for business trips the percentage share of bicycle trips reduces from 26.8% to about 11.8% in

13

14

extreme low and to 18.4% in low temperatures as compared to normal temperatures. Furthermore, the probabilities of car use increase by 12% and 6.3% for the same categories of temperatures. This reflects that in temperatures lower than 0o C there is a substantial change in percentage shares of business trips made by bicycle and car. Commuting trips also show comparatively large changes in percentage shares of bicycle and car trips in temperatures lower than 0o C. In contrast, the changes in percentage shares of bicycle and car trips for educational and recreational and sports trips are small in low temperatures. In extremely high temperatures, the probability of selecting the bicycle decreases substantially for business trips as compared to the other three types of trips. However, the general pattern of less percentage share of the bicycle trips and more percentage share of the car trips in total number of trips holds true for all the trip purposes in temperatures greater than 25o C. The marginal effects of public transport, walking and other are generally statistically insignificant or small. In general, the marginal effects of precipitation show the increasing probability of using the car and a decreasing probability of biking with increasing precipitation for all four trip purposes. The changes in percentage shares of the trips made by car and bicycle range from 1% to around 6 % for all type of trips.

4.

Conclusions and further research

This paper presents the influence of changing weather conditions on mode choice decisions of individuals in the Netherlands. We linked hourly weather data and transportation data, under the assumption that individuals base their mode choice decisions on the weather conditions that prevail during the hour of departure. We subsequently estimate a multinomial logit model to analyze the impact of weather on mode choice decisions. Furthermore, different models were estimated for different trip purposes. The results show that wind strongly discourages the use of the bicycle, but increases the propensity to use the car and public transport and to go walking. The effects of temperature show a decreasing percentage share of bicycle trips and an increasing percentage share of car and public transport trips at low temperatures. The opposite is true when temperatures increase up to 25o C, as compared to normal temperature. Once temperatures increase above 25o C, however, the probability of selecting the bicycle decreases and the probabilities of selecting the car and public transport increase. Finally, the effects of precipitation are strong for car and bicycle use. In addition, the probability of selecting the bicycle decreases and the probability of using the car and public transportation increases as the amount of precipitation increases. Similarly, the results of the models estimated for different trip purposes show similar patterns of changes in percentage shares of trips made by bicycle and car for all types of trip purposes. However, the magnitude of the effects is different for different trip purposes. A general conclusion is therefore that there is a substantial change in percentage shares of car and bicycle trips in changing weather conditions. This may imply that people prefer to bike in warm weather and under gentle wind conditions, whereas the car is preferred in opposite situations. Public transport, walking and other trips show changes in their percentage shares but these changes are small.

14

15

Our analysis shows the effects of weather conditions on individual mode choice decisions. This study does not answer whether the changes in percentage share of modes are due to substitution among travel modes or due to changes in overall travel demand. This issue deserves further research.

Acknowledgements We thank Nuffic, the Higher Education Commission of Pakistan and the Climate Changes and Spatial Planning program for sponsoring this research.

References Aaheim, H. A. and Hauge, E. K. (2005) Impacts of Climate Changes on Travel Habits, A National Assessment Based on Individual Choices. Center for International Climate and Environmental Research , Blindern, Oslo. http://www.cicero.uio.no/media/3816.pdf Andrey, Jean, Suggett, Jeff, Mills, Brian, and Leahy, Mike (2001) Weather-Related Road Accident Risks in Mid-Sized Canadian Cities. Canadian Multidisciplinary Road Safety Conference XII Proceedings, June 11-13, De Palma and Rochat (1999) Understanding Individual Travel Decision: Results from Commuters Survey in Geneva Transportation, vol. (26) pp. 263-281 Edwards (1996) Weather-Related Road Accidents in England and Wales: A Spatial Analysis Transport Geography, vol. (4) pp. 201-212 Eisenberg (2004) The Mixed Effects of Precipitation on Traffic Crashes Accident Analysis and Prevention, vol. (36) pp. 637-647 Hermans, Elke, Brijs, T., Stiers, T., and Offermans, Q. (2006) The Impact of Weather Conditions on Road Safety Investigated on an Hourly Basis. Transportation Research Board Annual Meeting 2006, CD-Room Paper Khattak and De Palma (1996) The Impact of Adverse Weather Conditions on the Propensity to Change Travel Decisions: A survey of Brussels Commuters Transport Resources, vol. (31) pp. 181-203 KNMI (2006) KNMI Climate Change Scenarios 2006 for the Netherlands.KNMI, De Bilt, Mandel, Gaudry and Rothengatter (1997) A Disaggregate Box-Cox Logit Mode Choice Model of Intercity Passenger Travel in Germany and its Implications for High-Speed Rail Demand Forecasts The Annals of Regional Science, vol. (31) pp. 99-120 McFadden (1974) The Measurement of Urban Travel Demand Journal of Public Economics, vol. (3) pp. 303-328

15

16

Oldenborgh and Ulden (2003) On the Relationship Between Global Warming, Local Warming in the Netherlands and Changes in Circulation in the 20th Century International Journal of Climatology, vol. (23) pp. 1711-1724 Stern and Zehavi (1990) Road Safety and Hot Weather: A Study in Applied Transport Geography Transactions of the Institute of British Geographers, vol. (15) no. 1, pp. 102-111 Stern, S. N. 2006, Economics of Climate Change Cambridge University Press, Cambridge. Suarez, Anderson, Mahal and Lakshmanan (2005) Impact of Flooding and Climate Changes on Urban Transportation: A Systemwide Performance Assessment of the Boston Metro Area Transportation Research Part D, vol. (10) pp. 231-244 Waalkes, S. M. (2003) Cold Weather and Concrete Pavements: Troubleshooting and Tips to Assure a Long-Life Pavement. Annual Conference of the Transportation Association of Canada St. John's, Newfoundland and LabradorLong-Life Pavements - Contributing to Canada's Infrastructure (A)" Session, Waerden, Peter van Der, Timmermans, H., and Borgers, Aloys (2002) Key Events and Critical Incidents Influencing Transport Mode Choice Switching Behavior: An Exploratory Study. Commission on Public Transportation Marketing and Fare Policy Winters, Friesen C. and Teschke K. (2007) Utilitarian Bicycling: A Multilevel Analysis of Climate and Personal Influences American Journal of Preventive Medicine, vol. (32) no. 1, pp. 52-58

16

17

Appendix A Coefficients of all variables full model (standard errors in brackets)* Bicycle Constant Distance Business Educational Recreational and Sports Income Urbanization Age02 Age03 Gender Car ownership Hour † Hour(10:00 to 17:00) Hour(17:00 to 19:00) Hour(19:00 to 07:00) Working day Summer Autumn Winter Wind † Wind greater than 6 BFT Temperature† Less than −8o C Between −8o to 0o C Between 10o to 20o C Between 20o to 25oC

Car

Bus, Tram, Subways, Train −1.5408 (.0454) 2.1477 (.0149) −.0389 (.0557) −.0188 (.0186) −.0346 (.0305) −.0254 (.0052) −.0819 (.0069) .3999 (.0239) .1014 (.0342) .0784 (.0175) −.1923 (.0220)

Other

.3759 (.0229) .9264 (.0229) −.0159 (.0261) −.0068 (.0089) .8055 (.0144) .0100 (.0024) .0797 (.0033) −.2166 (.0101) −.5379 (.0156) .0592 (.0083) −.2128 (.0126)

−2.3395 (.0254) 2.0601 (.0148) −.0461 (.0256) −.0048 (.0087) −.0122 (.0141) .0434 (.0022) .0625 (.0032) .5867 (.0102) .3610 (.0151) −.1296 (.0081) 1.5131 (.0163)

−1.8421 (.0511) 1.9615 (.0159) −.1002 (.0641) .0605 (.0209) .0303 (.0339) .0688 (.0052) .0562 (.0078) −.9957 (.0230) −1.4501 (.0413) −.4458 (.0194) −.4249 (.0288)

−.1420 (.0109) −.0376 (.0131) −.2789 (.0154) −.2168 (.0095) −.0013 (.0136) .0449 (.0117) −.0972 (.0142)

.1157 (.0109) .3585 (.0129) .3291 (.0144) −.0917 (.0092) −.0390 (.0136) .0363 (.0114) −.0142 (.0134)

−.7777 (.0208) −.5228 (.0247) −1.0308 (.0325) −.4098 (.0213) −.0956 (.0303) .1597 (.0245) .1222 (.0286)

−.0803 (.0251) .0111 (.0298) −.0397 (.0348) −.1681 (.0222) .0016 (.0305) −.0027 (.0268) −.1690 (.0344)

−.2128 (.0313)

−.0214 (.0294)

.1517 (.0587)

−.0103 (.0715)

−.4390 (.0432) −.2950 (.0189) .1271 (.0186) .1535

.0700 (.0392) .0416 (.0179) −.0285 (.0191) −.1984

.0578 (.0779) .0675 (.0375) −.0848 (.0451) −.2069

−.4848 (.1124) −.3437 (.0461) .0762 (.0413) .0068

17

18

Greater than 25o C Precipitation† Less then or equal to 1 mm

(.0280) −.1747 (.0116)

(.0300) .0643 (.0113)

(.0742) .0670 (.0243)

(.0644) −.1896 (.0268)

.0722 .0681 (.0330) (.0374) Greater than 1mm .1423 −.2262 (.0614) (.0788) Number of Observations 534079 Chi−Square 155118.5 Prob[Chi−square > value] 0.000 Loglikelihood −581237.3 Restricted Loglikelihood −658796.6 Pseudo R−Squared 0.1177 † Reference category for temperature, wind and precipitation are temperature between 0o and 10oC, wind up to 6 BFT, and no precipitation, respectively. Bold coefficients are statistically significant at 5%. −.1239 (.0164) −.1397 (.0317)

.0657 (.0154) .1410 (.0293)

18

19

Appendix B Model for Commuting Trips Table B1: Model fit information Number of Observations Chi−Square Prob[Chi−square > value] Loglikelihood Restricted Loglikelihood Pseudo R−Squared

79081 22474.88 0.000 −86311.87 −97549.31 0.1152

Table B2: Marginal effects (standard errors in brackets)*

Distance Income Urbanization Age02 Age03 Gender Car ownership

Hour † Hour(10:00 to 17:00) Hour(17:00 to 19:00) Hour(19:00 to 07:00)

Working day Summer Autumn Winter Wind † Wind greater than 6 BFT

Walk

Bicycle

Car

.2617 (.0039) .0068 (.0011) .0043 (.0016) .1437 (.0051) .1325 (.0078) −.0240 (.0040) .3492 (.0090)

Bus, Tram, Subways, Train .0222 (.0004) −.0012 (.0004) −.0043 (.0005) .0036 (.0018) .0014 (.0027) .0040 (.0013) −.0497 (.0018)

−.2250 (.0015) −.0047 (.0005) −.0099 (.0007) −.0256 (.0024) .0166 (.0035) .0053 (.0019) −.0966 (.0037)

−.0727 (.0031) −.0019 (.0009) .0092 (.0013) −.0947 (.0041) −.1167 (.0065) .0232 (.0033) −.1806 (.0066)

.0059 (.0025) −.0165 (.0030) .0086 (.0034)

Other

.0139 (.0005) .0011 (.0003) .0008 (.0005) −.0269 (.0014) −.0337 (.0027) −.0085 (.0012) −.0223 (.0019)

−.0306 (.0043) −.0465 (.0051) −.0781 (.0060)

.0518 (.0053) .0864 (.0061) .1063 (.0070)

−.0229 (.0015) −.0205 (.0018) −.0331 (.0025)

−.0043 (.0016) −.0030 (.0018) −.0038 (.0022)

.0230 (.0021) .0009 (.0031) −.0059 (.0026) .0012 (.0031)

−.0234 (.0038) .0017 (.0053) −.0017 (.0045) −.0061 (.0057)

.0102 (.00461) .0012 (.0065) .0029 (.0055) −.0003 (.0067)

−.0080 (.0016) −.0047 (.0024) .0040 (.0019) .0037 (.0022)

−.0018 (.0014) .0009 (.0019) .0008 (.0017) .0013 (.0022)

−.0142 (.0076)

−.0294 (.0133)

.0309 (.0156)

.0068 (.0049)

.0059 (.0044)

Temperature†

19

20

Less than −8o C Between −8o to 0o C Between 10o to 20o C Between 20o to 25oC Greater than 25o C

.0106 (.0090) .0234 (.0042) −.0143 (.0045) .0027 (.0069) .0066 (.0026)

−.1081 (.0168) −.0628 (.0075) .0166 (.0073) .0368 (.0114) −.0454 (.0045)

.0075 (.0056) .0054 (.0029) .0031 (.0033) .0007 (.0057) .0034 (.0018)

.0987 (.0191) .0446 (.0089) −.0095 (.0092) −.0479 (.0147) .0405 (.0055)

−.0088 (.0065) −.0106 (.0030) .0040 (.0025) .0076 (.0038) −.0051 (.0017)

Precipitation† Less then or equal to 1 mm

−.0001 .0012 .0147 −.0346 .0187 (.0036) (.0066) (.0077) (.0026) (.0023) Greater than 1mm .0053 −.0141 −.0398 .0583 −.0097 (.0071) (.0124) (.0143) (.0047) (.0055) † Reference category for temperature, wind and precipitation are temperature between 0o and 10oC, wind up to 6 BFT, and no precipitation, respectively.

Model for Recreational and Sports Trips Table B3: Model fit information Number of Observations Chi−Square Prob[Chi−square > value] Loglikelihood Restricted Loglikelihood Pseudo R−Squared

289823 84859.88 0.000 −315413.9 −357843.9 0.1185

Table B4: Marginal effects (standard errors in brackets)* Walk

Distance Income Urbanization Age02 Age03 Gender Car ownership

−.2319 (.0008) −.0039 (.0002) −.0101 (.0004) −.0252 (.0012) .0160 (.0018) .0065 (.0009) −.0876 (.0019)

Bicycle

−.0760 (.0016) −.0034 (.0004) .0088 (.0007) −.0949 (.0021) −.1297 (.0034) .0279 (.0017) −.1913 (.0034)

Car

.2696 (.0017) .0077 (.0005) .0054 (.0008) .1438 (.0026) .1442 (.0041) −.0296 (.0020) .3496 (.0047)

Bus, Tram, Subways, Train .0222 (.0002) −.0016 (.0002) −.0041 (.0002) .0055 (.0009) .0031 (.0014) .0045 (.0007) −.0488 (.0009)

Other

.0160 (.0002) .0011 (.0001) −.0001 (.0002) −.0292 (.0007) −.0336 (.0014) −.0094 (.0006) −.0219 (.0010)

20

21

Hour † Hour(10:00 to 17:00) Hour(17:00 to 19:00) Hour(19:00 to 07:00)

Working day Summer Autumn

Winter Wind † Wind greater than 6 BFT Temperature† Less than −8o C Between −8o to 0o C Between 10o to 20o C Between 20o to 25oC Greater than 25o C

.0039 (.0013) −.0257 (.0015) −.0040 (.0017)

−.0313 (.0022) −.0381 (.0026) −.0739 (.0031)

.0516 (.0028) .0866 (.0032) .1133 (.0037)

−.0234 (.0008) −.0208 (.0009) −.0337 (.0013)

−.0007 (.0008) −.0019 (.0010) −.0017 (.0012)

.0255 (.0011) .0023 (.0016) −.0063 (.0013)

−.0275 (.0020) .0053 (.0027) .0030 (.0023)

.0131 (.0024) −.0074 (.0034) .0007 (.0029)

−.0101 (.0008) −.0012 (.0012) .0041 (.0009)

−.0010 (.0007) .0009 (.0010) −.0014 (.0009)

.0096 (.0016)

−.0201 (.0029)

.0104 (.0035)

.0047 (.0011)

−.0046 (.0012)

.0137 (.0035)

−.0377 (.0067)

.0189 (.0078)

.0063 (.0023)

−.0011 (.0026)

.0307 (.0047) .0167 (.0021) −.0041 (.0022) .0073 (.0035) .0062 (.0013)

−.0716 (.0092) −.0532 (.0039) .0293 (.0037) .0481 (.0059) −.0335 (.0023)

.0480 (.0106) .0369 (.0046) −.0203 (.0048) −.0504 (.0077) .0296 (.0028)

.0048 (.0031) .0043 (.0015) −.0046 (.0018) .0061 (.0031) .0081 (.0009)

−.0119 (.0043) −.0047 (.0016) −.0003 (.0014) .0010 (.0022) −.0042 (.0009)

Precipitation† Less then or equal to 1 mm

.0032 −.0004 −.0305 .0247 .0029 (.0018) (.0034) (.0040) (.0013) (.0013) Greater than 1mm −.0017 .0018 −.0409 .0463 −.0055 (.0036) (.0065) (.0076) (.0025) (.0028) † Reference category for temperature, wind and precipitation are temperature between 0o and 10oC, wind up to 6 BFT, and no precipitation, respectively. * Bold coefficients are statistically significant at 5%.

21

22

Model for Business Trips Table B5: Model fit information Number of Observations Chi−Square Prob[Chi−square > value] Loglikelihood Restricted Loglikelihood Pseudo R−Squared

13409 4085.829 0.0000 −14502.08 −16545.00 .12348

Table B6: Marginal effects (standard errors in brackets)*

Distance Income Urbanization Age02 Age03 Gender Car ownership

Hour † Hour(10:00 to 17:00) Hour(17:00 to 19:00) Hour(19:00 to 07:00)

Working day Summer Autumn Winter

Wind † Wind greater than 6 BFT Temperature† Less than −8o C

Walk

Bicycle

Car

.2795 (.0084) .0074 (.0026) .0001 (.0040) .1583 (.0124) .1270 (.0193) −.0237 (.0098) .3507 (.0226)

Bus, Tram, Subways, Train .0233 (.0012) −.0012 (.0009) −.0023 (.0013) .0028 (.0044) .0030 (.0063) .0017 (.0033) −.0498 (.0045)

−.2841 (.0036) −.0053 (.0012) −.0123 (.0017) −.0324 (.0054) .0097 (.0078) .0120 (.0042) −.1055 (.0086)

−.0333 (.0083) −.0019 (.0023) .0128 (.0034) −.1049 (.0105) .1084 (.0166) .0250 (.0085) −.1704 (.0175)

−.0091 (.0056) −.0426 (.0069) −.0047 (.0075)

Other

.0146 (.0012) .0010 (.0007) .0017 (.0011) −.0238 (.0032) −.0312 (.0062) −.0150 (.0028) −.0250 (.0043)

−.0167 (.0114) −.0175 (.0132) −.0846 (.0160)

.0483 (.0133) .0784 (.0153) .1160 (.0176)

−.0213 (.0038) −.0174 (.0045) −.0268 (.0059)

−.0013 (.0037) −.0010 (.0043) .0001 (.0051)

.0153 (.0047) .0045 (.0069) −.0288 (.0060) −.0050 (.0071)

−.0104 (.0098) −.0110 (.0136) .0194 (.0117) −.0043 (.0152)

.0022 (.0112) .0054 (.0161) .0047 (.0137) .0051 (.0171)

−.0071 (.0039) .0035 (.0057) .0108 (.0046) .0070 (.0058)

.0001 (.0032) −.0023 (.0043) −.0060 (.0039) −.0028 (.0050)

.0486 (.0147)

−.0671 (.0337)

.0092 (.0373)

−.0039 (.0120)

.0131 (.0087)

.0368

−.1569

.1284

.0048

−.0131

22

23

Between −8o to 0o C Between 10o to 20o C Between 20o to 25oC Greater than 25o C Precipitation† Less then or equal to 1 mm Greater than 1mm

(.0199) .0224 (.0095) .0258 (.0094) .0199 (.0143) .0235 (.0059)

(.0473) −.0848 (.0201) .0372 (.0189) .0496 (.0285) −.0602 (.0118)

(.0502) .0633 (.0226) −.0660 (.0230) −.0775 (.0351) .0359 (.0136)

(.0152) .0075 (.0073) .0038 (.0079) .0029 (.0119) .0034 (.0046)

(.0176) −.0084 (.0069) −.0008 (.0062) .0052 (.0081) −.0025 (.0039)

−.0066 (.0080) .0190 (.0147)

−.0382 (.0165) −.0384 (.0313)

.0460 (.0185) .0276 (.0357)

.0035 (.0061) .0080 (.0102)

−.0047 (.0058) −.0162 (.0137)



Reference category for temperature, wind and precipitation are temperature between 0o and 10oC, wind up to 6 BFT, and no precipitation, respectively. * Bold coefficients are statistically significant at 5%.

Model for Educational Trips Table B7: Model fit information Number of Observations Chi−Square Prob[Chi−square > value] Loglikelihood Restricted Loglikelihood Pseudo R−Squared

54721 15428.09 0.0000 −59734.29 −67448.33 0.1143

Table B8: Marginal effects of variables (standard errors in brackets)* Distance Income Urbanization Age02 Age03 Gender Car ownership

Walk

Bicycle

Car .2505 (.0041) .0076 (.0013) .0057 (.0019) .1465 (.0062) .1476 (.0094) −.0312 (.0048) .3304 (.0107)

Bus, Tram, Subways, Train .0203 (.0005) −.0021 (.0005) −.0037 (.0006) .0090 (.0022) .0029 (.0032) .0012 (.0016) −.0463 (.0021)

−.2048 (.0021) −.0031 (.0007) −.0073 (.0009) −.0278 (.0031) .0127 (.0045) .0084 (.0024) −.0861 (.0046)

−.0800 (.0038) −.0030 (.0011) .0054 (.0016) −.1013 (.0050) −.1289 (.0078) .0312 (.0039) −.1736 (.0077)

.0055

−.0251

Other .0140 (.0006) .0006 (.0004) −.0002 (.0006) −.0265 (.0017) −.0343 (.0032) −.0096 (.0014) −.0244 (.0022)

.0429

.0231

−.0003

Hour † Hour(10:00 to 17:00)

23

24

(.0032) −.0114 (.0038) .0061 (.0043)

(.0051) −.0371 (.0061) −.0786 (.0072)

(.0064) .0759 (.0074) .1081 (.0084)

(.0018) .0222 (.0022) −.0325 (.0029)

(.0019) −.0051 (.0023) −.0031 (.0027)

.0233 (.0027) .0057 (.0040) −.0035 (.0034) .0118 (.0041)

−.0277 (.0046) .0098 (.0063) .0110 (.0055) −.0198 (.0068)

.0108 (.0055) −.0152 (.0078) −.0153 (.0067) .0045 (.0081)

−.0051 (.0018) −.0020 (.0027) .0035 (.0022) .0063 (.0026)

−.0013 (.0017) .0016 (.0024) .0043 (.0204) −.0028 (.0026)

Wind greater than 6 BFT Temperature†

.0264 (.0083)

−.0580 (.0150)

.0184 (.0174)

.0098 (.0047)

.0034 (.0049)

Less than −8o C

.0366 (.0118) .0102 (.0054) −.0034 (.0056) .0116 (.0089) .0038 (.0034)

−.0690 (.0213) −.0376 (.0091) .0224 (.0087) .0802 (.0137) −.0350 (.0053)

.0354 (.0247) .0358 (.0108) −.0136 (.1109) −.0795 (.0185) .0286 (.0067)

−.0003 (.0072) −.0013 (.0034) −.0087 (.0044) −.0103 (.0077) .0018 (.0022)

−.0026 (.0081) −.0071 (.0037) .0032 (.0031) −.0019 (.0055) .0008 (.0020)

Hour(17:00 to 19:00) Hour(19:00 to 07:00)

Working day Summer Autumn Winter Wind †

Between −8o to 0o C Between 10o to 20o C Between 20o to 25oC Greater than 25o C Precipitation† Less then or equal to 1 mm Greater than 1mm

.0048 −.0022 .0116 −.0348 .0207 (.0046) (.0078) (.0092) (.0029) (.0029) .0116 .0069 −.0057 −.0436 .0309 (.0086) (.0148) (.0172) (.0050) (.0058) † Reference category for temperature, wind and precipitation are temperature between 0o and 10oC, wind up to 6 BFT, and no precipitation, respectively. * Bold coefficients are statistically significant at 5%.

24

Suggest Documents