Evidence for prospect-refuge theory: a meta-analysis of the findings of environmental preference research

In the modern world, designers and planning authorities are responsible for creating and approving new spaces that contribute to a pleasant, healthy and sustainable environment. As such, policies, standards and procedures for architectural, urban and landscape designers increasingly anticipate the use of approaches which have a positive impact on human perceptions and behaviour (Shaftoe 2008; Gutman 2009). Such expectations have, since the 1970s, been driving an increased practical interest in explaining or predicting human responses to space and form (Perloff 2015). Indeed, theories about the way in which people perceive and respond to the spaces they inhabit can be found in the oldest architectural treatises and the earliest utopian works (Kruft 1994). Not surprisingly then, the concept that a particular combination of space, form and context might have a positive effect on a person’s emotional state is often raised in the design disciplines (Menin 2003; Crankshaw 2008) and it has even found its way into various guidelines and primers (Lidwell et al. 2003; Lippmann 2010).

Probably the most known theory for explaining environmental preference in the architectural, interior and urban design disciplines is ‘prospect-refuge theory’, although its application in design actually combines aspects of Berlyne’s (1951) ‘arousal theory’ and Kaplan and Kaplan’s (1989) ‘information model’ with Appleton (1975) prospect-refuge theory. The merging of these different theories, along with several additional themes, was proposed by Grant Hildebrand (1991), an architectural historian attempting to explain the innate appeal of several of Frank Lloyd Wright’s residences. Hildebrand’s (1991, 1999) research combined several different explanations of human perceptions and emotions to propose a formula for creating an ideal environment.

In the last two decades this expanded variation of prospect-refuge theory has seemingly become accepted in the design disciplines as offering an explanation of basic human responses to the environment, and as a type of guideline for creating ideal spaces (Kellert 2005; Lippmann 2010). However, Hildebrand’s argument is entirely qualitative, as too is Appleton’s, which is also, despite the way it is cited in architecture, about preferences for landscape views. Nevertheless, architectural applications of this theory are allegedly seen in the works of renowned designers including Alvar Aalto and Pritzker prize winners Glenn Murcutt, Jørn Utzon and Peter Zumthor (Gallagher 2007). But once again, the evidence for these claims is qualitative and circumstantial. Indeed, arguments for the efficacy of prospect-refuge theory in design only rarely identify specific quantitative or empirical studies to support their claims. This situation is the catalyst for the present paper, which revisits the key quantitative evidence that is available for prospect-refuge theory (and its associated theories) to collectively assess their findings and relevance to the design of landscapes, cities, buildings and interiors.

Originally, developed by Appleton (1975) for explaining preferences for certain landscapes, prospect-refuge theory argues that we derive feelings of safety and pleasure from inhabiting environments that offer both views and a sense of enclosure. This appeal is arguably universal and subconsciously influences our day-to-day decision-making (Ellard 2009). The central assumptions of prospect-refuge theory can be connected to both Darwinian nineteenth century anthropological beliefs about survival instincts as well as to phenomenological studies that examine environments in relation to the human body (Husserl 1973). Prospect-refuge theory also has parallels with arousal theory, which suggests that an increase of pleasure is felt when a person views a space or scene that has a degree of uncertainty or novelty about it, but if uncertainty is increased beyond that point, feelings of anxiety begin to occur (Berlyne 1951).

In architecture and design, Hildebrand (1991, 1999) expanded the standard definition of prospect-refuge theory to include four additional spatio-cognitive elements: mystery, complexity, enticement and illumination. Most of these can be traced to Kaplan and Kaplan’s (1989) information theory framework which suggests that environments that provide increased opportunities for gathering or discovering information allow for improved living conditions including heightened safety. Thus various spatio-cognitive properties associated with exploration potential (including complexity and mystery) also have an impact on environmental preference.

As a result of the work of Hildebrand (1991), a growing number of architectural critics and historians have used an expanded definition of prospect-refuge theory to justify or analyse architectural spaces in terms of psychological wellbeing (Jacobsen et al. 2002; Roberts 2003; Gallagher 2007; Unwin 2010). The resultant model of a preferred environment has four components. First, a space must have a view or outlook, and second, that view must be at least partially framed or enclosed. Third, a degree of visual complexity enhances feelings of safety and finally, a sense of mystery (implying discoverability or directionality) is preferred (Dosen and Ostwald 2013a).

Despite this apparent agreement between researchers, the capacity to understand and shape human perceptions of environments is actually a more complex and vexed topic. For example, the way people perceive environments is clearly different depending on their physical stature and size or the personal experiences (Wohlwill 1976; Heerwagen 1998). Despite this, architectural and design texts tend to selectively emphasise the universal aspects of spatial experience (Kellert 2005; Lippmann 2010). This may not be entirely unreasonable because people do experience aspects of space, including its organisation, in similar ways (Ellard 2009). For example, quantitative studies have shown that a close visual connection between habitable space and nature is beneficial for psychological wellbeing, recovery and stress relief (Kaplan and Kaplan 1989; Heerwagen and Orians 1993). Studies have also observed that restricted views may cause negative reactions (Heerwagen 2008) while visual connections might encourage movement and evoke pleasure through the exploration of space (Kaplan 1987). There is evidence that people have a preference for spaces which feature access to nature and daylight (Ulrich 1993) and for surroundings that support social interaction and thereby create safer environments (Kuo and Sullivan 2001; Ellard 2015). In addition, a preference for water views, which are perceived as calming, is often noted in studies (Ulrich 1984; Kaplan and Kaplan 1989; Heerwagen and Orians 1993). However, Stamps (2006, 2008a, b) examined ratings for comfort and preference for both natural and built environments and concluded that the statistical significance of prospect, refuge and luminosity factors, in shaping emotional response, was ‘very near zero’ (Stamps 2008b, p 141). Indeed, the only factor which Stamps found had any evidence of shaping environmental preference was ‘venue’, being the particular type of environment where the studies were undertaken. Stamps’ findings emphasise the fact that the evidence for environmental preference theory, and especially as it relates to different types of designed environments—the urban, architectural and interior—is more complex than it seems.

The present paper summarises and classifies the results of thirty-four studies that have used quantitative means (generally data derived from surveys, interviews or computational and mathematical analysis) to examine the veracity of four specific spatio-cognitive factors—prospect, refuge, mystery and complexity—as part of spatial preference theory in design. The first two of these factors are the most commonly tested, as they were the earliest proposed by Appleton (1975), while research into the latter pair is less common, and is often associated with Hildebrand’s (1991, 1999) identification of them as being equally significant in architecture and design. Importantly, this paper does not undertake a detailed review of the methods that have been used to test these factors, as the majority have already been examined from a methodological perspective (Dosen and Ostwald 2013b). Instead, the results of the thirty-four studies are each classified in terms of whether their findings support the efficacy of each of the four factors, or are neutral, or contrary in terms of their role shaping environmental preference.

In addition, to reviewing the results of the thirty-four studies holistically, they are also divided into three groups in accordance with their specific focus or venue of testing. The three groups are: urban studies (9 %), territorial studies (29 %), and architectural studies (62 %). This division by venue or focus is significant as it assists in differentiating which evidence can most reasonably be used to support arguments about interiors, cities and landscapes. Furthermore, while findings associated with the four specific factors—prospect, refuge, mystery and complexity—will be collectively considered in the conclusion, many results are specific to the environment or venue they were tested in. For example, the ratings for prospect and refuge conditions can be heavily influenced by concerns about safety when being exposed in a natural or urban environment, while this is less of a concern when being inside a building. Thus, the results of the various studies analysed here are often valid in a particular context, but not necessarily in any other.

The thirty-four quantitative studies chosen for this meta-analysis comprise the full set of works that fulfil five criteria. First, they have a focus on testing at least one of the following environmental characteristics: prospect, refuge, mystery and/or complexity. Second, these studies use quantitative methods. Third, they are readily accessible, meaning they are either published or available through databases. Fourth, they are cited in the architecture and design literature as evidence for the efficacy of the four perceptual properties, or use keywords that are directly linked to these properties. Finally, all of the studies offer some conclusions, based on the evidence they have collected, about the validity of the factor or factors being considered. Importantly, for each of these thirty-four studies only their findings are compared in this paper, not their methods. Furthermore, no direct commentary is provided here about the statistical rigour or empirical validity of their results, although in all cases it could be said that the results appear reasonable, given the methods used and their stated or otherwise obvious practical limitations. Nevertheless, the paper does examine some of the demographic information which is recorded in past research to assist with interpreting the findings and directions for future research.

Because environmental preference theory combines both spatial and psychological properties, the thirty-four studies include those focussed largely on perceptual properties (twenty studies), those focussed on geometric properties (five studies) and those which combine the two (nine studies). The perceptual properties are generally tested using surveys whereas all but one of the geometric studies use isovist analysis. For this reason, it is useful to define each of the four major properties which are compared in the present paper in both perceptual and geometric terms. First, prospect is associated with the perceptual properties of outlook, depth of view, spaciousness and openness, and the geometric properties of isovist area and maximum radial line length. Second, refuge is associated with the perceptual properties of enclosure and safety, and the spatio-visual geometry of occlusion and minimum radial length. Mystery relates to transitions between spaces, to changing luminosity and the degree of occlusivity. Complexity refers to the volume of information present in a space, along with the number of occluding edges (or vertices) and the jaggedness of its geometry.

When assessing the thirty-four studies, all were initially classified in three ways—supportive, neutral and contrary—in respect of claims about preferences for prospect, refuge, mystery and complexity. In general, findings directly relating to these spatio-cognitive properties were accepted prima-facie. That is, the author’s conclusions were accepted in the majority of cases. However, where a study only indirectly tested one of the spatio-cognitive properties, it was classified in accordance with the pertinent section of the data. For example, if the primary purpose of the study was to determine if men and women have similar spatial perceptions and its secondary purpose was to see if they have similar tastes, then the result of the gender differentiation part of the test was ignored in our classification.

Findings that confirmed the basic assumptions of the expanded variation of prospect-refuge theory were classified as ‘supportive’ (indicated with a ‘√’ in the tables). However, it is important to remember that not all positive findings of this type were equally emphatic and, as the tabulated summaries and quotes reveal, many were only partially supportive. Neutral findings (indicated with a ‘O’ in the tables) are ones wherein the evidence is either not statistically significant, or it is only pertinent to testing a meta-principal, like the consistency of an argument or of a spatial property. In the former case, partially supportive findings were categorised in the neutral group if they relied on less than 66 % of the data or equivalent confidence indicators. In the latter case, many of the neutral studies used computational or mathematical approaches to test the presence or absence of a claimed prospect-refuge pattern in spatial data. Thus, some of these computational studies may confirm the presence of a pattern, but if it does not clearly conform to the expectations of prospect-refuge theory, it is classed as neutral here. Contrary findings (indicated with a ‘X’ in the tables) are examples of the failure to find evidence to support the impact of one of the factors, or of clear evidence against the significance of one or more factors.

Altogether, findings of thirty-four studies are summarised in this paper of which three are related to urban environments, ten examine landscapes and twenty-one investigate interiors. Most of the studies involving landscapes and urban environments were conducted in the 1980s and 1990s (eight out of thirteen) while interior studies were generally more recent, including all of the mathematical-computational studies or those using combined approaches (Fig. 4).

While acknowledging that the veracity of each set of results is most strongly pertinent within its venue grouping, considering the complete set of results is also informative. With thirty-six supported results for factors, thirty-four neutral and twelve contrary, the evidence is mixed. The complete set of results (Table 4) shows that the most examined characteristic is, not surprisingly, prospect (38 %), and it is also the most confirmed (53 %). Prospect is a dominant topic of testing especially in interior studies where it is excluded in only two out of twenty-one tests, and it is also the lead factor tested in landscape studies. Refuge, an indicator for safety and privacy, is the second most supported factor (22 %) but it has the highest rate of both neutral (41 %) and contrary findings (41.5 %). Prospect has the second highest percentage of all neutral findings (29 %) of which most belong to the results for interiors (8 out of 10). Certainly, very few studies present results that are completely contrary (14 %), of which most reject, as just indicated, refuge (41.5 %), followed by mystery (33 %) and then prospect (17 %) and complexity (8.5 %) (Fig. 5). Mystery and complexity are the least examined characteristics (14.5 % each) and have been tested predominantly in interior settings. Within its category, complexity has more supportive results (six) than neutral (five) or contrary (one) (Fig. 6). In total, when examining the results of all of the thirty-four studies holistically in this way, there are not enough supportive findings to emphatically substantiate prospect-refuge theory.

Number of studies that are	P	R	M	C	SUM
Supportive √	19	8	3	6	36
Neutral O	10	14	5	5	34
Contrary X	2	5	4	1	12
SUM	31	27	12	12

P prospect, R refuge, M mystery, C complexity

If the results are divided by venue, then the most supportive findings associated with prospect are related to interior venues (Fig. 7). Nine out of twenty-one studies confirm that wide, open rooms are perceived as more comfortable than enclosed ones. Curiously, only one of these interior studies presents a preference for a balance of openness and enclosure, while in landscape studies three out of ten confirm a preference for balanced settings. According to Scott (1993a), “people like vertically and horizontally expansive settings that are subdivided into smaller spatial zones” (13), a result that also confirms a need for enclosure in interiors. Originally, Appleton (1975) proposed that environments which offer a balance of both outlook and enclosure would be most preferred. In a later revision he advises that in many cases a “balance […] can be achieved from serial vision, involving the successive experiences of exposure to strongly contrasting landscape types, strong prospect and then strong refuge, is more potent than that which comes from trying to achieve a balance all at once” (Appleton 1984, p 102). This suggestion, which was developed to describe landscape preference, may also relate to the experience of interiors, as appealing designs were often composed of large, clearly open and smaller, enclosed spaces rather than offering both in one room. Perhaps because of this uncertainty, if there is need for enclosure or not, refuge is the least supported characteristic within the interior group (and also overall) with twelve (out of seventeen) neutral findings.

A preference for complexity has been confirmed by four interior studies which used combined approaches. Also, two survey-based studies confirm a preference for spaces that have more complex geometry (Scott 1993a, b). In addition, a second test also presents a preference for spaces that permit internal views and which offer natural content and daylight. Altogether, this is the only study in the entire set that confirms the importance of all four factors (Scott 1993a).

It is notable, when considering the studies compiled and examined here, that the majority of the survey results involved relatively small numbers of participants. Only eight of the 29 studies engaged 100 or more participants while 16 studies relied on responses of 20 or less participants. That such small studies remain widely cited as evidence of the efficacy of prospect-refuge theory is concerning. Furthermore, the larger studies tended to have more neutral results, leading to additional concerns about the usefulness of some past research (Nasar et al. 1983; Kaplan and Herbert 1988; Nasar 1988; Loewen et al. 1993). To further interpret the results, the demographic data associated with survey participants can be considered. For example, only six studies recruited participants of varying background and ages whereas most of the studies relied on surveys of university students. Two studies (Conrad 1993) engaged participants who were trained in design as well as others who had no training in design to compare their responses, while another two studies explicitly selected only participants that were untrained (Scott 1993a, b). Only two studies developed a cross-cultural comparison, however, one of these was of participants from North-America and Australia; groups with relatively similar living conditions and cultural values (Nasar 1988; Kaplan and Herbert 1988). Furthermore—with one exception (Nasar 1988)—most of the studies are also of Western cultures. None of the studies considered the local background of participants—meaning the area, for example, an urban, a suburban or rural environment, where a participant has predominantly lived—which may have a strong influence on spatial preference.

Problematically, in terms of research design, some of the stimuli sets used for surveys did not possess clear or carefully calibrated or graduated differences. Controlled changes of one or more variables are necessary in stimuli when accuracy is required, and most of the more recent studies test virtual environments, which allow such controlled variations. However, few of the early studies used such controlled conditions, leading to difficulties replicating or understanding their results. This is a problem because some studies have shown that the background of a stimuli can have a significant influence on the ratings of an image (Stamps 2008a, b).

Overall, only a few studies produced statistically thorough results and analysis. Interestingly, some of these were undertaken by the same researcher and contradict his own previous findings, thereby demonstrating the difficulty of producing an emphatic outcome. For example, Stamps (2008a) finds that an increase of refuge is associated with an increase of feelings of comfort, but his following study presents contrary results (2008b). Also one of the interior studies that used a combined approach contradicts an earlier finding as isovist area (openness) did not correlate any more with the rated factor of beauty but only with spaciousness, complexity and sociability (Dzebic 2013). Another (neutrally categorised) finding is unexpected, as more complex spaces (which can be related to a low isovist-area-to-perimeter ratio) seem to allow a higher recognition rate of words and images that had to be memorised in such complex interiors while the area (large or small) made no difference on such ratings (Dalton et al. 2010).

Significantly, many of the studies that have been categorised as neutral are based on mathematical or computational analysis of interiors testing more than one measure. This often leads to discrepancies in their results. So far, there is no clear determination of which measures are the most relevant, but there are assumptions that, for example, in isovist analysis (which twelve of the interior studies use), ratios or proportional values relate better to spatial dimensions than pure measures (Dawes and Ostwald 2014a, b). Such works are part of the recent trend to examine prospect-refuge theory mathematically or in a combined approach with survey methods and computational analysis. However, there is still need to refine these new research methods by implementing clear, controlled changes and seeking more detailed correlations between human perceptions and spatio-geometric measures.

Despite what seems to be widespread acceptance in the architectural, urban and landscape design fields, the quantitative evidence for prospect-refuge theory remains inconsistent. It is especially problematic that the results which are most commonly cited in architecture relate to studies of natural environments, not interiors or urban environments. As this paper demonstrates, the results are most valid in specific venues. For example, the benefits of a close visual connection to nature and of inhabiting a space that offers both an open area for outlook and a more private area for being hidden, have been broadly supported by past research in natural settings. However, the same spatio-visual configuration (the same volume of outlook and enclosure) in an interior overlooking a city skyline, will trigger a different psychological reaction. In particular, enclosure appears to be primarily significant in natural and urban settings. These two results are especially challenging for architectural arguments about the need for prospect and refuge, as the evidence for the latter is far weaker than for the former, and especially so for interiors. More specifically for designers, the results for complexity seem to confirm that a degree of complexity in interior space is preferred, but they are unclear about how much or where it should be. The results for mystery are less emphatic (75 % neutral or negative) although few of the spaces tested in these studies appear to possess a high level of mystery. Moreover, very few of the hypothesised triggers for mystery have ever been examined using quantitative means. For example, visual connections, changing levels of light and varying ceiling heights (Hildebrand 1991, 1999) have been claimed to be significant components of preferred spaces, but only two of the studies examined here consider them and both are neutral in their findings.

Of equal importance to the meta-analysis and results contained in this paper is their implications for future research. As identified in the previous discussion section (Table 5), there are clear messages in this paper about the type of research that is needed to clarify the conflicting messages found in past studies, and fill substantial gaps in our knowledge. In particular, future research should test in more detail which degree of prospect and refuge and also mystery would be perceived as comfortable in a natural or an urban context. Additional criteria including daylight, illumination and the venue itself—for example, is it a relatively safe area, is it a crime-intense area or are there any other hazards?—would need to be considered. Also, the relevance of refuge in such a context would be interesting to examine. Which degree of prospect and refuge would achieve aesthetic quality in a natural or urban setting? Furthermore, it might be wise to test interiors that depict habitable space and which consist of more than one room, as previous studies considered mostly external views from one room but not so often internal view connections between spaces. This may change the results for refuge as well as for mystery and may identify a need for both.

Ultimately, the gaps in environmental preference research for architects, urban and landscape designers are both significant and largely unacknowledged outside small groups or researchers. For professionals and students, prospect-refuge theory seems to offer a simple explanation or formula for creating space, even though, as this paper demonstrates, the unanswered questions are more significant than the answered ones, or alternatively that the unknowns far exceed the knowns. These gaps have to be closed if designers are to use evidence-based strategies to shape our landscapes, cities and buildings.