EMPIRICAL EVIDENCE
VALUING TIME SAVINGS
The fundamental problem facing the revealed preference approach in practice is the type of bias we discussed in Chapter 3. When doing revealed preference analysis, the treatment is a good with a certain attribute (such as being only 10 minutes from the city), while the control is another good without that attribute (such as being 5 minutes farther from the city). The problem is that the treatments and controls may differ in ways that lead to bias. Suppose that homes built closer to the city are smaller, or that they have smaller yards. This would lead their value to be lower, so that when one compared the prices of houses farther away and closer to the city, one might not find the expected decline in prices for farther -away homes. In the Boston metropolitan area, for example, the town of Everett is on average only 4 miles from downtown Boston, while the suburban town of Lexington is 11 miles away. Yet the average home price in Everett is $322,923, while the average home price in Lexington is about 2.3 times higher at $746,804.9 This is because the houses in Lexington are typically much larger and have nicer attributes than those in Everett. Many of these attributes are observable, such as the square footage of the house or the number of bathrooms. In such cases, we can try to control for these other attribute differences using cross -sectional regression analysis with control variables. Indeed, in this context there is a name for such a strategy: hedonic market analysis. Hedonic market analysis proceeds by running a regression of house values on each of the bundle of attributes of housing: distance to town center, number of bedrooms, number of bathrooms, square footage, and so on. The notion is that if we control in a regression context for all of the attributes other than distance, we will essentially be comparing identical houses in different locations. As we highlighted in Chapter 3, however, this is not likely to be a fully satisfactory approach. There are many differences between houses that are hard to observe, such as the perceived quality of the neighborhood or the care taken by the previous owner. If these things are correlated with distance to the town center, it will mean that the treatment group (close houses) and the control group (more distant houses) are not identical products, so our (biased) estimates do not give a true valuation of time differences. In order to provide a more convincing estimate of the value of time savings, a quasi -experimental approach can be used. An example of such a study was done by Deacon and Sonstelie (1985). During the oil crisis of the 1970s, the government imposed price ceilings on the large gasoline companies, setting a maximum price that those companies could charge per gallon of gas. These low prices (relative to the true market price) led consumers to wait in long lines to get gas. These price ceilings did not apply to smaller, independently owned stations, so lines were shorter there. As a result, the amount of time individuals were willing to wait at the stations owned by large gas companies (the treatment group) relative to independent stations (the control group) can be compared to the amount of money saved by going to the treatment stations instead of the control stations to form a value of time. The authors compared Chevron stations in California, which were mandated to lower their prices by $0.45 per gallon (in 2009 dollars) below the price being charged by the control group of independent stations. Lines formed at Chevron stations for cheaper gas, forcing customers to wait an average of 14.6 minutes more there than at competing stations. The mean purchase was 10.5 gallons, suggesting roughly that people were saving $19.00 (in 2009 dollars) per hour they waited. That is, individuals revealed themselves to be willing to wait an hour for $19.00—almost exactly equal to the average hourly wage in the United States.10 E
หลักฐานเชิงประจักษ์ประเมินค่าประหยัดเวลาThe fundamental problem facing the revealed preference approach in practice is the type of bias we discussed in Chapter 3. When doing revealed preference analysis, the treatment is a good with a certain attribute (such as being only 10 minutes from the city), while the control is another good without that attribute (such as being 5 minutes farther from the city). The problem is that the treatments and controls may differ in ways that lead to bias. Suppose that homes built closer to the city are smaller, or that they have smaller yards. This would lead their value to be lower, so that when one compared the prices of houses farther away and closer to the city, one might not find the expected decline in prices for farther -away homes. In the Boston metropolitan area, for example, the town of Everett is on average only 4 miles from downtown Boston, while the suburban town of Lexington is 11 miles away. Yet the average home price in Everett is $322,923, while the average home price in Lexington is about 2.3 times higher at $746,804.9 This is because the houses in Lexington are typically much larger and have nicer attributes than those in Everett. Many of these attributes are observable, such as the square footage of the house or the number of bathrooms. In such cases, we can try to control for these other attribute differences using cross -sectional regression analysis with control variables. Indeed, in this context there is a name for such a strategy: hedonic market analysis. Hedonic market analysis proceeds by running a regression of house values on each of the bundle of attributes of housing: distance to town center, number of bedrooms, number of bathrooms, square footage, and so on. The notion is that if we control in a regression context for all of the attributes other than distance, we will essentially be comparing identical houses in different locations. As we highlighted in Chapter 3, however, this is not likely to be a fully satisfactory approach. There are many differences between houses that are hard to observe, such as the perceived quality of the neighborhood or the care taken by the previous owner. If these things are correlated with distance to the town center, it will mean that the treatment group (close houses) and the control group (more distant houses) are not identical products, so our (biased) estimates do not give a true valuation of time differences. In order to provide a more convincing estimate of the value of time savings, a quasi -experimental approach can be used. An example of such a study was done by Deacon and Sonstelie (1985). During the oil crisis of the 1970s, the government imposed price ceilings on the large gasoline companies, setting a maximum price that those companies could charge per gallon of gas. These low prices (relative to the true market price) led consumers to wait in long lines to get gas. These price ceilings did not apply to smaller, independently owned stations, so lines were shorter there. As a result, the amount of time individuals were willing to wait at the stations owned by large gas companies (the treatment group) relative to independent stations (the control group) can be compared to the amount of money saved by going to the treatment stations instead of the control stations to form a value of time. The authors compared Chevron stations in California, which were mandated to lower their prices by $0.45 per gallon (in 2009 dollars) below the price being charged by the control group of independent stations. Lines formed at Chevron stations for cheaper gas, forcing customers to wait an average of 14.6 minutes more there than at competing stations. The mean purchase was 10.5 gallons, suggesting roughly that people were saving $19.00 (in 2009 dollars) per hour they waited. That is, individuals revealed themselves to be willing to wait an hour for $19.00—almost exactly equal to the average hourly wage in the United States.10 E
การแปล กรุณารอสักครู่..
