Sample Nonresponse in a Mail Contingent Valuation Survey: An Empirical Test of the Effect on Value Inference

Abstract

The contingent valuation method (CVM) is often applied for valuing nonmarket environmental resources (Cummings, Brookshire, & Shulze, 1986; Johansson, 1987; Mitchell & Carson, 1989). This method uses constructed market scenarios to elicit people's valuations in terms of, for example, their willingness-to-pay (WTP) for a proposed environmental change. The scenario may be presented to the individuals in connection with an interview investigation or a mail survey. In recent years, the mail survey technique has grown in popularity due to its being cheaper and avoiding interviewer bias.A mail CVM survey is generally confined to a sample of individuals, which is chosen to be representative of a specific population. However, for various reasons there are always some individuals who fail to complete and return the questionnaire they receive by mail. As a consequence, the number of observations becomes less than the number of individuals in the original sample. Then, the problem is whether the WTP derived from the realized sample (the respondents) adequately represent the WTP held by the original sample. In other words, what effect does the sample nonresponse have on value inference for the population?Many factors may contribute to sample nonresponse. Some individuals may fail to answer because of less interest in the survey object. In this case, it is plausible to assume that the nonrespondents would have a lower WTP than the respondents. However, other individuals may fail to complete and return the questionnaire due to such reasons as lack of time, reluctance to be involved in any survey, and limitations of the abilities to understand the questionnaire and the constructed market scenario (Carson, 1991). Since there is no a priori knowledge about what reasons underlie the nonresponse in a specific survey, it should not be taken for granted that the value inference from the realized sample is representative.The Right of Common Access in Sweden allows people to enter any forest land, no matter who owns the land. As a result, people spend a fair amount of their leisure time in forests pursuing activities such as hiking, camping, wildlife watching, berry- and mushroom-picking. To assess the value of forests as environments for these activities, a mail CVM survey was undertaken. This paper reports a test of whether environmental value inference from this survey is representative for the whole sample (and for the population), or, in other words, whether sample nonresponse bias occurs.METHODOLOGYA CVM survey of residents of Vasterbotten county in northern Sweden (where forests cover about 70% of the total land area) was undertaken in the autumn of 1991. The questionnaire was sent to a random sample of 2000 individuals, and 1245 answers were returned (one reminder was required to obtain 430 of them). Questions concerned forest recreation habits, personal characteristics, and preferences for alternative forest landscapes (Mattsson & Li, in press). The individuals were urged to think over, among other things, the implications of their access to and experience of the forest environment. Then, they were confronted with the discrete choice CVM question: Imagine that it would be associated with a cost to enjoy the forest environment-Would you be willing to pay $ x in order to continue to visit, use, and experience the forest environment as you usually do! (where x was a specific bid, in SEK, varied across the individuals in the sample), the answer being yes or no.When all returns were recorded, a follow-up questionnaire was sent to a sub-sample of 500 individuals in the original sample; 259 were respondents and 241 were nonrespondents to the CVM survey (the nonrespondents included both item-nonrespondents to the discrete choice question and nonrespondents to the whole CVM questionnaire). The sub-sample was selected by choosing every fourth person from the list of individuals in the original sample. …