Explaining landscape preference heterogeneity using machine learning-based survey analysis

Abstract

We conducted a national survey on a high-quality internet panel to study landscape preferences in Norway, using photos as stimuli. We examined preference heterogeneity with respect to socio-demographic characteristics and latent topics brought up by the respondents, using ordinal logistic regression and structural topic modelling (STM), a machine learning-based analysis. We found that pasture landscapes are the most favoured (55%), while densely planted spruce forests are the least favoured (8%). The contrast was particularly strong between eastern and western Norway, between men and women, and between young and old. STM revealed that the choices were mainly driven by the preference for landscape openness, especially by women. Other important drivers were concerns regarding reforestation of former farmlands, aesthetic properties, forest management, biodiversity issues, and cultural values. Our results suggest that landscape policies may clash with socio-cultural preferences, and failure to account for these may undermine the success of a policy.

How Do Observable Characteristics of Post-Mining Forests Affect Their Attractiveness for Recreation?

“…Gatersleben and Andrews [9] found that natural environments with a higher level of prospect (clear field of vision) and lower level of refuge are more restorative than environments with a lower prospect and high refuge. The positive impact of an open view has been corroborated by yet other authors, e.g., [12,13,24]. Their findings imply that less permeable and potentially dangerous environments demand more attention, lead to a lower restorative potential, and may be, therefore, less preferred.…”

“…For example, the volume of large trees increases the preference for the forest more in men than in women [22]. Older women in Norway preferred grassland and heathland (compared to forests) more than other population groups [24]. Ribe [25] shows that people favoring resource production have lower standards of visual quality/scenic beauty than people favoring resource protection.…”

“…Mean tree height affects the preference more in a group of forest owners than in a group of non-owners [22]. Different explanations exist for heterogeneity in preferences towards the same forest structures: the effect of familiarity on the environmental preference [11,24]; different underlying values besides the aesthetics such as ecological values, attitudes, and interests [25][26][27], background knowledge, and available information [16,19,25]. In effect, different factors of preference may be identified in studies conducted on different types of samples, such as experts in forestry, interested groups, and landowners, e.g., [14][15][16]25] or the general public, e.g., [16] and visitors, e.g., [13].…”

“…The quotas used for the sampling were general socio-demographic characteristics that are commonly used for quasi-representative samples of the population [40]. At the same time, region, age, gender, and education of people have been found to explain differences in environmental preference, e.g., [12,22,24] and, hence, might have been confounding variables in the model.…”

A one-hundred-year structural topic modeling analysis of the knowledge structure of international management research

“…Major topics detected include value for money, baggage policy, delays and refund policy. Liu et al ( 2021 ) used STM to analyze responses to 1452 open-ended questions in a survey examining landscape preferences in Norway. Results revealed that consumer choices were driven mainly by themes such as landscape openness, concerns regarding deforestation, biodiversity issues and aesthetic properties.…”