The visual impact of agricultural sheds on rural landscapes: The willingness to pay for mitigation solutions and treatment effects

•We assessed residents’ WTP to mitigate the visual impact of sheds in UNESCO rural areas.•WTPs for the same solution differ by the type of visual information provided.•We presented different econometric approaches to estimate treatment effects.•The theoretical equivalence of a fully interacted with split models holds in CL, not in MXL models.•Monetary valuation can support the design of appropriate landscape management plans. New rural and industrial sheds cause significant transformations determining a negative change in landscape perception. In this paper the willingness to pay (WTP) of residents to mitigate the visual impact of sheds in the rural areas of the UNESCO World Heritage Site “Vineyard Landscape of Piedmont: Langhe-Roero and Monferrato” is estimated through a Choice Experiment (CE). The CE survey was carried out through face-to-face interviews involving 400 residents. The set of mitigation attributes was selected by means of a landscape analysis and presented to respondents using a set of images (close and distant views). Conditional Logit (CLogit) and Mixed Logit (MXL) models both in preference and in WTP space were estimated. From the empirical point of view, the study shows that sheds are generally perceived as negative landscape elements, and that residents are willing to pay for mitigation solutions, such as tree lines and formal hedgerows. The results are relevant for policy makers since they suggest how agricultural sheds are perceived and which mitigation strategies are preferred by local residents. From the methodological point of view, we estimate and test different models for assessing the effect of the visual treatment (close vs distant view). We show the theoretical equivalence of a fully interacted model (i.e., estimated on the whole sample and including interaction terms with the treatment), with split models of the choice for separate subsamples submitted or not to the treatment. We find that in estimation the equivalence holds in the results of Conditional Logit models, but not in Mixed Logit models. The reasons and implications are discussed.