Choice experiment surveys have been used for many years in transport economics and market research and are now becoming increasingly popular for the valuation of environmental and health-related goods. They present sampled respondents with different choice sets, each comprising a finite set of alternatives defined on a number of attribute dimensions, and require respondents to specify their preferred alternative in each choice situation. Each alternative involves a bid amount to be paid, which generally equals zero for the Status Quo (SQ), as well as the level of each relevant non-monetary attribute of the good. The respondent’s task is to state his/her preferred option (Hensher et al. 2005). This setting is consistent with the random utility maximization model and several econometric treatments have been developed to analyze data from choice experiments. More information on choice experiment and the specifications of our model is available as Electronic Supplementary Material.
Study Area and Scenario Development
Lodz is the third largest city in Poland and it is the city with the smallest area of streetside greenery among major Polish cities. The numbers of urban trees in general and street trees in particular have been decreasing in Lodz (and in most Polish cities) in recent years and the living conditions for trees have worsened (Kronenberg 2012; Kabisch and Haase 2013). There is no inventory of street trees in the city, and management practices in the city center are largely restricted to removing trees that are judged “in bad condition” or because of new infrastructural developments. Compensation plantings are ordered only when healthy trees are removed and they mostly take place outside of the center.
Lodz, and especially its center, is widely perceived as gray and neglected, and it is suffering from unsatisfactory environmental health indicators. For example, in 2003 Lodz had the highest mortality rate due to respiratory diseases of both men and women among all Polish cities with more than 100 000 inhabitants (Wcisło 2008).
Our study focused on the city center within which the total length of streets is about 50 km. This is a densely built area. Many streets are lined with narrow strips of unpaved ground that used to be green, with lawns and trees, but from which the trees have been removed over time without ever being replaced. For the purposes of this study, we performed a rough analysis of streets in the city center, classifying them into four categories: (i) “High”—streets with many trees (10 or more trees per 100 m), currently 12 km; (ii) “Medium”—streets with medium number of trees (4–9 trees per 100 m), currently 10 km; (iii) “Islets”—streets with trees planted on islets, currently 0 km; and (iv) “No trees”—streets with no or few trees (0–3 trees per 100 m), currently 28 km.
After consultation with landscape specialists it turned out that, in the most optimistic scenario, the following improvements are possible in terms of planting trees: (a) upgrading a maximum of 8 km of streets from “Medium” to “High”; (b) upgrading a maximum of 20 km of streets from “No trees” to “Medium”; and (c) for 8 km of streets, it is not possible to plant enough trees to change their “No trees” status. The improvement from “No trees” to “Medium” can be achieved either by planting trees in the space between sidewalk and road (there is enough space for planting additional trees in this way along maximum 8 km of streets) or by creating islands in the parking places or on the road, which may be possible on maximum 12 km of streets. Table 1 presents the attributes and their levels used at the designing stage.
The choice sets were generated following the Street et al. (2005) and Street and Burgess (2007) optimal-in-difference design approach. Each respondent was faced with 12 choice situations, involving the choice between the SQ alternative, with no tree planting program and no payment required, and three program alternatives. Respondents were asked to select the best alternative in each of 12 choice sets.
To make things simpler, in the questionnaire we translated the attributes and levels from design stage into following categories: (i) Length of streets with a high number of trees; (ii) Length of streets with a medium number of trees; (iii) Length of streets with islets; and (iv) Length of streets with no trees.
The payment vehicle used in the survey was monthly increase in the local tax that all Lodz citizens would have to pay.Footnote 1 An example of a choice card is presented in Fig. 1.
The survey was conducted between July and November 2011. The questionnaire was administered face-to-face on a sample of the Lodz population, with interviews conducted in public places. Questionnaires were randomly assigned to 400 individuals and 351 valid questionnaires were collected and used in the econometric analysis described in this paper.
The questionnaire consisted of four parts. The first one included questions about respondent’s attitude toward trees in the city. The second part described the current situation in Lodz, using maps (Fig. 2) and photos to illustrate the attributes and their levels. The third part of the survey was the choice experiment. The forth part contained debriefing questions and collected socio-economic data, including gender, age, location, education, household characteristics, and income.
Two models were estimated on the data. We begin with a basic Multinomial Logit (MNL) model, with no random preference heterogeneity (model 1). This is then followed by a second model, which allows for random preference heterogeneity with correlation between individual coefficients, Mixed Multinomial Logit (MMNL).
The utility for the SQ alternative is given by a constant. The utility function for the three program alternatives includes continuous coefficients associated with:
High, upgrade from “Medium” to “High”;
Medium, upgrade from “No trees” to “Medium”;
Islets, upgrade from “No trees” to “Islets”;
and cost of the program.
In the MNL model, in addition to the main effects, we included four interactions of non-monetary attributes with the following socio-demographic variables: age, gender, education, and car ownership. By adding into the utility function the cost/income ratio we also allowed the cost sensitivity to vary with income level.
Finally, in all estimated models, we used a linear specification of attributes of the utility functions. This is based on preliminary analyses that did not reveal consistent and significant non-linearities in response with the data at hand. The modeling results are presented in Table 2. All models were coded and estimated in Nlogit 5.0.