Show someone two photographs of a street and ask which they prefer. The answer may depend on the light, camera angle, visible people, and a guess about what lies beyond the frame. It may tell us little about which street they would choose to use at night, live beside, or defend from redevelopment. Environmental preference is a judgment about settings under a particular presentation and question. Measuring it well requires an explicit choice about the target: visual liking, anticipated use, desire to explore, or preference among real options. The same scene can rank differently when those questions change.
Define the judgment and comparison
Preference can be measured as a rating of one environment, a ranking of several, or a choice between pairs. Each method creates a different comparison. A high rating says that a scene meets a respondent’s standard under that scale; a forced choice only says that it won against the alternatives offered. If every option is unattractive, a winner need not be positively valued. If the alternatives differ in cost, distance, safety, and access, a choice can express tradeoffs rather than scenic liking. Researchers should state the decision the participant was asked to imagine before interpreting the result.
The environmental object also needs boundaries. A photograph represents a view; a park is a lived setting with routes, noise, weather, social use, and rules. Research titled “Picture or Place?” demonstrated that evaluations of photographs and of the places represented can differ. A photograph-based experiment can isolate visual features and efficiently present many scenes, but it should claim visual preference under the display conditions. An on-site study adds movement and sensory context, though it brings variation in weather and activity. Neither method is automatically superior; each supports a different inference.
Preference is distinct from place attachment and place identity. People may prefer a new, attractive garden over a familiar square while remaining strongly attached to the square. An identity-bearing setting may be uncomfortable yet central to a person’s history. A visual rating cannot be used to infer a bond or a willingness to remain. Conversely, attachment can shape preference, especially when the participant recognizes the place. Researchers can measure these constructs separately and examine their relation, but merging them in one score makes both difficult to understand.
Preference is also distinct from perceived restoration. Hartig and colleagues developed a measure of restorative quality that asks about features thought to support recovery of attention. A person can find a scene attractive without believing it would be restorative, and a restorative setting need not be the most visually exciting. Using a restoration scale as if it were a preference scale changes the construct. If a study wants both outcomes, it should measure and report both, then test whether they coincide for the people and environments studied.
What scene ratings can test
Kaplan, Kaplan, and Brown compared domains of predictors of environmental preference across scenes and found that different kinds of attributes offered different explanatory value in their study context. Their work is a reminder that preference is not captured by one physical property. Perceptual qualities, informational features, land-cover categories, and content can all matter. A model that performs well in one landscape type should not be treated as a universal equation for public taste. To test a proposed predictor, researchers need scenes that vary sufficiently on that predictor without confounding it completely with other salient qualities.
Herzog’s study of urban nature examined content categories, viewing time, and several candidate predictors, including coherence, mystery, and nature. The result illustrates how controlled scene ratings can explore a cognitive account of liking. It does not establish that these features determine every person’s everyday use of a space. Viewing time and the mix of scenes are features of the experiment. If all coherent scenes are also brighter, greener, or more familiar, statistical coefficients may not isolate coherence itself. Researchers should describe stimulus selection and use design or analysis to address such confounding.
Stamps synthesized research on the informational variables coherence, legibility, complexity, and mystery. The synthesis is valuable because it tests whether a favored framework holds across studies rather than relying on one visually persuasive example. It also shows why measurement quality matters: different raters, scenes, and methods can alter estimated relations. An informational variable is often itself judged by participants or experts, not read directly from the image. If the same respondents rate both mystery and preference, shared response style can inflate their association. Independent ratings can reduce one concern but introduce others, such as disagreement between experts and users.
A strong stimulus set covers the intended universe of environments. Ten carefully chosen photographs of tidy green streets cannot estimate preference for all urban settings. If photographs come from one season, city, or camera perspective, the study may mainly test those choices. Random or stratified sampling of scenes can improve generalizability. When complete sampling is impossible, researchers should describe the range of attributes and contexts represented. This is analogous to sampling people: scene selection determines the population of environments to which a result can reasonably apply.
Standardizing images can help isolate a feature but can also erase the very context people use in a real decision. Equalizing brightness, crop, or vegetation may make scenes more comparable visually while removing cues to maintenance or safety. Digital edits can create implausible combinations. An experiment should report exactly what was manipulated and validate whether viewers notice and interpret the change as intended. A preference difference between edited images is evidence about that manipulation under those conditions. It is not proof that the same intervention in a lived street would produce the same response.
People do not all see the same environment
Personal history, cultural learning, mobility, and current purpose affect a preference judgment. A winding path may signal exploration to one person and difficulty navigating to another. A crowded square may feel lively or unsafe depending on experience and time of day. Cross-cultural work comparing responses to landscape images has challenged universal claims about preferred openness. Such studies do not imply that every individual difference is arbitrary; they show that a sample’s background is part of the inference. Reporting recruitment and relevant experience is essential when recommendations will affect diverse users.
Accessibility changes what an environment affords. A scenic staircase may receive high visual ratings but be unusable for someone who cannot climb it. A narrow trail may feel inviting in an image while excluding people using mobility devices. If the study asks what respondents would choose to visit, anticipated access belongs in the construct. If it asks for visual beauty only, accessibility should not be inferred from the answer. Researchers can gather both ratings and analyze where they diverge. This distinction prevents design decisions from favoring an attractive picture over a place that people can actually use.
Familiarity and recognition can also change the task. Residents who know a pictured place may incorporate memories, social meaning, or concern about future development, while outsiders evaluate the visible scene alone. A mixed sample should record recognition and consider separate analyses. Hiding place labels can reduce some cues but cannot prevent all recognition. Showing labels may be appropriate when the target is an actual planning decision. The key is to be explicit about whether the judgment is meant to concern anonymous visual form or a named place with history.
Social desirability may enter when an option carries moral meaning. A respondent may endorse a natural landscape because they believe they ought to care for nature, even if they would choose a shaded urban plaza for everyday use. Anonymity, clear wording, and behavioral tradeoff questions can help reveal differences between stated ideal and practical selection. Yet no single response is the true preference in every context. Preference is conditional on goals and constraints. A good study reports those conditions rather than trying to locate a context-free ranking inside the participant.
Analysis and validity checks
Likert ratings are simple but vulnerable to differences in scale use. Some respondents avoid endpoints; others use them readily. Pairwise comparisons can reduce reliance on a common numeric standard, but the number and order of pairs can make the task burdensome. Choice experiments can estimate tradeoffs among designed attributes, yet their conclusions depend on the attributes and levels offered. Repeated ratings allow assessment of consistency, but perfect consistency is not a criterion of validity if the scenes are ambiguous or the participant’s goals change. Select the response format to fit the decision rather than because it produces the neatest coefficient.
Order and context effects deserve attention. A modest park may seem attractive after several traffic-heavy scenes and ordinary after an exceptional garden. Randomizing presentation order helps estimate such effects, though randomization does not fix unrepresentative scene selection. Short breaks can reduce fatigue when many images are shown. Reporting response time and missing ratings can reveal whether the task became perfunctory. If a participant sees a scene for only a moment, the resulting judgment may rely on color and composition more than on inferred wayfinding or possible use.
Construct validation asks whether the measure behaves as the theory predicts without collapsing into adjacent constructs. For example, if a study says it measures desire to explore, a forced choice between “walk into this scene” and “avoid it” may fit better than a generic beauty rating. The investigator can also ask why a scene was selected, using open answers to test whether the intended feature was noticed. Independent assessments of scene attributes can reduce common-method bias. Convergent evidence is useful, but disagreement may show that beauty, safety, restoration, and use are meaningfully different outcomes.
Prediction of actual behavior is another question. A stated preference for a park may be consistent with later visitation, but travel time, opening hours, cost, weather, social company, and accessibility intervene. A field study can track use, yet use itself is not a pure preference measure: people may frequent the only available space. A design that combines stated ratings with observed opportunity and behavior is more informative than treating either as a perfect criterion. If the aim is a planning forecast, disclose the practical constraints included and those left out.
Interpretation for design and policy
An average preference score can hide polarized judgments. A design liked moderately by everyone may have the same mean as a design loved by half and avoided by half. Distribution matters, especially if a change burdens a group with less influence. Report subgroup patterns with care, avoiding stereotypes or overinterpretation of small samples. Qualitative explanations can show whether disagreement concerns safety, memory, aesthetics, function, or trust in management. Public decisions need these reasons, not just an aggregate ranking of images.
Preference is an input to design, not a decision rule. Ecological function, accessibility, heritage, housing, maintenance, and procedural fairness may matter even when they are not visible in a scene rating. A picture of mature trees can receive a high score while the proposed design is impractical under local conditions; a less immediately attractive intervention may bring long-term benefit. Researchers can inform deliberation by making the preference evidence precise. They should not use a visual survey to claim that the public has approved every consequence of a project.
Jayakody and colleagues’ review of place attachment and climate adaptation provides context for this distinction. It reports that bonds with place can support or constrain action depending on risk perception, values, and local processes. That is not evidence that an environmental preference rating predicts adaptation. A person may prefer a proposed greener landscape yet oppose a plan that displaces neighbors. A study of preference and policy support should measure both and investigate the reasons for divergence. The concepts become useful when their boundaries remain visible.
What the evidence does not show
Scene ratings do not establish a universal human preference for one landscape form, prove that pictured features caused a judgment, or predict use without considering context. A photograph cannot capture every sound, smell, route, social interaction, or change over time. A statistically reliable preference scale does not automatically measure restoration, place attachment, identity, or wellbeing. The reviewed studies support specific methods and conditional associations, not an algorithm that can select the best environment for everyone. Claims should remain attached to the participants, stimuli, response task, and decision context actually studied.
Environmental preference becomes more useful when the study tells us precisely what people chose between, what they thought the choice meant, and whose alternatives were represented. Ratings can reveal patterns that designers would otherwise miss; interviews and field observation can reveal why those patterns may change once a picture becomes a place. The measure’s value lies in opening a more informed conversation about environments people can inhabit, not in compressing every judgment about a setting into the number assigned to its image.
Research references
- Kaplan, R., Kaplan, S., & Brown, T. (1989). Environmental preference: A comparison of four domains of predictors. Environment and Behavior, 21(5), 509–530. View source
- Herzog, T. R. (1989). A cognitive analysis of preference for urban nature. Journal of Environmental Psychology, 9(1), 27–43. View source
- Stamps, A. E. (2004). Mystery, complexity, legibility and coherence: A meta-analysis. Journal of Environmental Psychology, 24(1), 1–16. View source
- Scott, M. J., & Canter, D. V. (1997). Picture or place? A multiple sorting study of landscape. Journal of Environmental Psychology, 17(4), 263–281. View source
- Hartig, T., Korpela, K., Evans, G. W., & Gärling, T. (1997). A measure of restorative quality in environments. Scandinavian Housing and Planning Research, 14(4), 175–194. View source
- Hägerhäll, C. M., Ode Sang, Å., Englund, J.-E., Ahlner, F., Rybka, K., Huber, J., & Burenhult, N. (2018). Do humans really prefer semi-open natural landscapes? A cross-cultural reappraisal. Frontiers in Psychology, 9, 822. View source
- Jayakody, D. Y., Adams, V. M., Pecl, G., & Lester, E. (2024). When does place attachment lead to climate change adaptation and when does it not? A quantitative review and thematic analysis. Environmental Science & Policy, 160, 103866. View source