I read that contingent valuation surveys can only provide estimates of WTP for discrete environmental changes. Conversely, choice experiments can be used to infer marginal WTPs. Is that generally true?
This is the kind of question that will split the research environment. I'd say that first of all it's kind of the wrong question, because the simple answer is 'yes' but since the question is wrong, that's not much help. A survey/experiment will always provide an estimate. The question is whether that estimate is any use. A better question is 'can CV provide VALID and ROBUST estimates of WTP?' In that sense the answer is much more likely no. Largely because for 'evironmental changes' the nature of the change is so underdetermined for most purposes in the data you get back will both have a very high variance and very limited validity - the actual environmental change is likely to be experienced (and valued) in very different ways to which the CV estimate is generated. Consequently, you will get estimates - you always do with a survey - but whether they tell you anything you should act on is a different matter.
The same is true with choice experiments since the cases offered in the experiment will very likely (unless you are willing/able to do a very complex design with very realistic conditions) suffer from the same issues as in a CV.
These issues aren't really specific to CV studies their true of any study aiming to get some sense of future preferences or attitudes or prospective behaviour dependent on some described future conditions.
Thanks all. Somehow every time I ask a question at RG, I start a deep conceptual debate, which is not what I intended... I am well aware of the limitations of CV. My question is of a technical nature: can CV provide estimates of marginal willingness-to-pay or is it restricted (as I read in a book) to discrete, non-marginal WTP? That's all I need. Thanks for reading suggestions about the limitations of economic valuation methods, but believe me - I know about most of them.
For marginal WTP choice experiment will be a good option. You can use NLOGIT or STATA for this. In STATA you can use mixlogit command for estimation and then follow the wtp. both are user written command. Follow ssc install mixlogit