Identifying the Effects of Low Emission Zones, ch. 3-4

3. Theoretical Analysis

Following the analytical framework developed by Schmutzler (2011) , a policy directed towards reducing total emissions from road transportation can operate through different channels: through the car kilometers travelled, via a change in the specific emissions of cars, via a change in public transport kilometers travelled, through a change in the specific emissions of public transport, and lastly through the effect on emissions from general economic activities. Specific emissions of cars and public transport are defined as the emissions under consideration (in our case, PM10) per kilometer travelled. Low emission zones like the ones implemented in Germany are likely to affect at least three of the five channels named above.

First of all, people whose car does not fulfill the standard could decide to use other means of transportation, thereby reducing the number of car kilometers travelled. Probably, this reduction would increase the kilometers travelled by public transport. Yet , as long as the specific emissions of public transport are lower than the ones of passenger cars, total emissions are going to fall. Another possibility is to buy a new car that fulfills the emission standard or to do a retrofit on the old one . Both reactions decrease the specific emissions of passenger cars per kilometer. However , the investment in a new and probably more fuel efficient car could cause a so-called rebound effect: By decreasing the private costs per kilometer travelled, it is possible that the number or length of car journeys increases (Litman, 2005). Hence , the effect on total emissions of buying a new car is unclear. The fourth channel, specific emission of public transport, is unlikely to be affected by LEZs as they are designed in Germany7. General equilibrium effects, such as the relocation of businesses following the implementation of an LEZ, are possible but hard to predict.

The framework can also be used to analyze effects on regions adjacent to the one  where a policy is introduced. The number of car kilometers travelled can either increase or decrease following policy implementation: If the total number of trips is reduced, then car kilometers travelled might also decrease in areas close to the LEZ , as well as public transport kilometers might increase. Yet , if people choose to relocate activities, such as shopping or leisure, to places outside the LEZ or decide to drive around the restricted area to reach a certain destination, then car kilometers travelled outside the zone might increase. If a rebound effect is present, it is likely to affect places outside the LEZ as well. A change in specific emissions, however, is likely to decrease even the emissions outside the LEZ.

From this analysis, it can be seen that the both direct effect of LEZs and possible spillover effects depend on the behavioral responses of people to the policy. Especially the presence of a rebound effect could offset some of the emission reductions achieved through the other channels, emphasizing the need for empirical evaluations of LEZs. The following sections will introduce the data used and outline the empirical strategy.

4. Data and sample selection

Data on PM10 concentrations was obtained from the Federal Environmental Agency on a daily basis for the years 2004 – 2010, including the observation date, the daily mean PM10 concentration, and station characteristics such as the measuring site (urban, rural, etc.) and the type of station (measuring background, traffic or industry pollution). The time span 2004 until 2010 was selected because the first air quality plans entered into force in 2004 and the last year for which validated data was available was 2010. As the dataset included all German measuring stations that were active in one of those years , a selection of the relevant stations for both the treatment and the control group had to be done first. The selection of treatment and control observations was based on a list of 125 cities with an air quality plan in place during that time span, provided by the Federal Environmental Agency (2012c).

The treatment group was created by first selecting all cities that implemented an LEZ between 2004 and 2010 into the sample, resulting in 39 cities. Following that, all stations measuring PM10 pollution at traffic sites in these cities were determined, using the public station database of the Federal Environmental Agency (2012d). While in some cities several stations measuring PM10 at traffic sites are / werewere available, other cities were not equipped with such a station at all. [/annotax] Such cities were deleted from the sample, while in cases of several stations per city, all were kept. In the next step, the data availability was checked, since not all stations were active in the whole period 2004 – 2010. All stations that contained data for at least two successive years were kept in the sample. For example, if data for a station was available for 2008 and 2009; it was kept, while a station with data for 2005 and 2007 but not 2006 was deleted. Dropping these units was necessary as it am / is / waswas unclear why[/annotax] such missing years occurred and how they affect data quality. In some cases, the reason for disruptions was reported in the data base, such as maintenance work or a change in the measuring technique. Howeverthat was not the case for all stations, and since it is unclear if such disruptions systematically affect data quality between the years, stations with missing years were dropped.

For the control group, all cities that established an air quality plan but not an LEZ between 2004 and 2010 were selected into the sample. The establishment of an AQP within two years is required as soon as a violation of one of the EU limit values is recorded. Thus , the control group does not represent a random sample of cities without an LEZ, but a selected group that experienced problems with PM10 limit values before. While sampling randomly from all German cities to generate the control group could result in systematic differences in PM10 concentration trends between the groups, all cities implementing a / anan AQP are undertaking some measures to reduce their PM10 concentrations.[/annotax] As the similarity of underlying trends is a crucial assumption for the empirical strategy to be used, the focus on AQP cities is the most suitable approach. As in the case of treated stations, some control cities or stations had to be dropped from the sample due to missing data.

It have to / have got tohas to be noted that[/annotax] selecting cities based on the implementation of AQPs and LEZs means that the sample selection is ultimately based on the outcome variable, PM10 concentrations. If group assignment is determined by unobserved characteristics that also influence the outcome variable, this could threaten the assumption of exogeneous treatment assignment, which is crucial for a valid estimation of the causal effect. However , as will be further explained in sections 5 and 6, such unobserved characteristics can assumed to be time-invariant in the sample selected. Such time-invariant unobserved characteristics can be handled if panel data is available, which is the case in this study (see for example Meyer 1995).

After deleting cities without an appropriate measuring station and dropping those with missing data, the sample contained 122 measuring stations. 42 of them were treated with a low emission zone step 1 at some point between 2004 and 2010. Ten out of these 42 successively introduced a low emission zone step 2 and four out of the 42 an LEZ step 3. These stations were summarized to 14 units treated with an “advanced” low emission zone. The final sample contained thus 80 units in 62 different cities that only had an air quality plan in place, 28 units in 21 different cities that introduced an LEZ step 1 (from now on: basic LEZ) and 14 units in 6 different cities that first introduced an LEZ step 1 and later an LEZ step 2 or 3 (from now on: advanced LEZ). In total, the 122 cross-sectional units produced 260,423 data points on the mean daily concentration of PM10, which  were converted into monthly averages, resulting in 8556 observations. Of these , 236 (2.76%) observations are missing. The share of missing values is slightly higher for the group of stations only treated with an air quality plan (3.25%), while for the groups treated with a basic or advanced LEZ, 1.70% respectively 1.99% of all observations are missing. There is no large difference in the share of missing values when comparing the periods before and after treatment implementation: For the cities treated with a basic LEZ, 1.58% of PM10 data is missing before treatment implementation and 1.94% after it, while the shares are 1.8% respectively 2.28% for the cities treated with a basic LEZ.

Table 3 provides descriptive statistics on some characteristics of the three different groups. Average PM10 concentrations as well as the number of days in exceedance of the 24 hour limit are higher in treated than in control cities, and again higher in cities introducing an advanced LEZ than in those introducing only a basic LEZ8. PM10 background concentrations, which measure the general urban PM pollution, are also slightly higher in treated cities, which could indicate that LEZs are more likely to be introduced in generally more polluted cities. Quite similar for all three groups is the environment the measuring stations are placed in; the large majority of stations are placed in an urban setting.

Table 3: Descriptive statistics on the characteristics of the different treatment groups (2004-2010) [Table not shown]

An important aspect that cannot be seen from table 3 is that LEZs are distributed spatially imbalanced within Germany. Of the 42 observations having an LEZ in place, 23 are located in the federal state of Baden-Wuerttemberg in the southwest of Germany. The remaining 19 LEZs are spread out over 6 different states9, resulting in 9 out of 16 federal states with only AQPs in place. This requires a strategy that can account for developments relevant to both treatment assignment and outcomes arising from this imbalance. Factors influencing PM concentrations that could otherwise be considered random, such as meteorological conditions or economic shocks, could be concentrated in states where either a large share of the treatment or the control observations are located, and hence be correlated with both the treatment status and the outcome (see Auffhammer et al. (2009) for a discussion of spatial aspects in evaluating PM10 pollution control policies). To control for potential economic shocks, data on employment, disposable income and population was collected from the Federal Statistics Agency, but such data was only available until 200910.

To control for variations in the microclimate between stations, preferentially data capturing meteorological influences should be used, such as on precipitation, wind speed, and temperature. However , the collection of such detailed data was not possible within this study. The issue of spatial distribution is therefore handled by fixed effects (see the empirical strategy section), but the collection of suitable control data for weather conditions should be taken into consideration in further research.


7  Air quality plans, in contrast , often include the decrease of PM10 emissions from public transport, most often from buses, as a measure to be taken.

8 More detailed statistics on the distribution of the yearly average PM10 concentrations and the number of days in exceedance of the 24h limit can be found in the appendix.

Six are located in Northrhine-Westphalia, 5 in Bavaria, 3 in Berlin, 1 in Bremen, 2 in Hessia, and 2 in Lower Saxony.

10 Due to the limited date availability, the economic control variables will only be used for robustness checks on a subsample of the original data, including only the observations for 2004–2009 .