Norm Crystallization: Measurement and Comparative Analysis

Abstract

Normative theory and methods have become increasingly important in outdoor recreation research and management as an approach to formulating standards of quality in parks and related areas. An important issue associated with this research is the level of agreement or consensus about social norms called crystallization. A new measure, Van der Eijk's measure of agreement (A), is proposed and applied in this study. A test of the effect of potential independent variables on crystallization is also applied using a comparative analysis conducted across a number of studies administered in U.S. national parks. Study findings indicate a generally high level of crystallization regarding normative standards of quality and show several independent variables that influence crystallization, though the effects are generally small.

Keywords:

• crystallization
• national parks
• normative theory and methods
• standards of quality
• Van der Eijk's measure of agreement (A)

Several park and outdoor recreation management frameworks have been developed in the professional literature and applied by public agencies and related organizations. Prominent examples include Limits of Acceptable Change (LAC) developed by the U.S. Forest Service (Stankey et al., 1985) and Visitor Experience and Resource Protection (VERP) developed by the U.S. National Park Service (Manning, 2001; National Park Service, 1997). Although some differences exist among these and related frameworks, most are founded on the concept of indicators and standards of quality (Manning, 2004). Indicators of quality are measurable manageable variables that serve as proxies for management objectives, including the desired level of resource protection and the type and quality of the recreation experience to be provided and maintained. Standards of quality define the minimum acceptable condition of indicator variables. Once management objectives and associated indicators and standards of quality are formulated, indicators are monitored and management actions taken to ensure that standards of quality are maintained.

One of the challenging elements of this management approach is formulating standards of quality. Standards of quality can be informed through many sources, including legal and policy mandates, historic precedent, interest group dynamics, and public opinion, especially that derived from park and recreation visitors. This latter source has special appeal because it involves people most directly interested in and affected by park and outdoor recreation management policy.

Normative Theory and Methods

Norms are a theoretical construct that have a long tradition and are widely used in sociology and the social sciences more broadly (Vaske & Whittaker, 2004). Norms represent what is “normal” or generally accepted within a cultural context (Manning, 2007). Normative theory and related empirical methods have increasingly been used to help formulate standards of quality in parks and outdoor recreation and broader natural resources contexts (Donnelly et al., 2000; Manning, 2007; Shelby & Heberlein, 1986; Shelby, Vaske, & Donnelly, 1996; Vaske & Donnelly, 2002; Vaske & Whittaker, 2004). If visitors and other stakeholders have norms concerning relevant aspects of park and outdoor recreation conditions and experiences, then such norms can be measured and used as a basis for formulating standards of quality.

Application of normative theory and methods to parks and outdoor recreation has raised several conceptual and empirical issues (Heywood, 1993a, 1993b, 1996a, 1996b; Manning et al., 2001; McDonald, 1996; Noe, 1992; Patterson & Hammitt, 1990; Roggenbuck et al., 1991; Shelby et al., 1996; Williams, Roggenbuck & Bange, 1991). As originally conceived, norms are understood to (a) guide behavior (Biddle, 1986; Blake & Davis, 1964; Cancian, 1975), (b) be enforced by sanctions (Homans, 1950; Rossi & Berk, 1985) and (c) be widely shared by social groups (Rossi & Berk, 1985). Although these underlying assumptions may not be met explicitly in the context of parks and outdoor recreation, researchers (e.g., Manning, Johnson & Vande Kamp, 1996; Shelby & Vaske, 1991) have argued that (a) outdoor recreation often involves emerging norms for which a strong sense of obligation and related sanctions are evolving, (b) outdoor recreation-related norms can apply to social and environmental conditions as well as behavior because such conditions are often a direct outcome of behavior, (c) outdoor recreation-related norms often regulate collective rather than individual behavior (i.e., agency-imposed regulations on levels and types of use) and (d) outdoor recreation research has begun to document some degree of consensus about a number of recreation-related norms.

FIGURE 1 Hypothetical social norm curve (Manning, 1999).

Normative research in parks and outdoor recreation has been guided primarily by structural characteristics models of normative theory (Vaske & Whittaker, 2004). In particular, Jackson's (1965) “return potential model” works by asking survey respondents (e.g., park visitors, community residents) to evaluate the acceptability (or other evaluative dimension as described by Manning et al., 1999) of a range of visitor-caused impacts to park resources or the quality of the visitor experience. For example, visitors to a wilderness area might be asked to rate the acceptability of encountering a range of groups of other hikers along a trail per day. Resulting data (i.e., personal norms) are averaged and plotted as shown in Figure 1 to construct a social norm curve. Such findings can help guide formulation of standards of quality. For example, the point at which the social norm curve crosses the neutral point of the response scale might be considered the minimum acceptable condition. This point is when the number of groups of hikers encountered is judged by the sample as a whole to fall out of the acceptable range and into the unacceptable range (though it may be reasonable to consider some other point along the social norm curve when formulating standards of quality based on consideration of management objectives, resource sensitivity or other matters). Application of normative research to standards of quality in parks and outdoor recreation has been described and compiled in several sources (i.e., Manning, 1999; Manning, 2007; Shelby & Heberlein, 1986; Shelby et al., 1996; Vaske et al., 1986). This body of work has often focused on the issue of crowding but has been extended to other park and outdoor recreation-related issues, as well as to concerns associated with broader environmental and natural resources management (Vaske & Whittaker, 2004; Manning, 2007).

Norm Crystallization

As described, the degree of consensus about recreation-related norms is an important issue. This issue is illustrated graphically in Figure 1 and is represented by the dispersion of data around the points defining the social norm curve, which is often called “crystallization” and was identified by Jackson (1965) as an important property of social norms. If there is reasonable agreement about norms within a stakeholder group, then normative data can be useful in guiding formulation of standards of quality.

Concern over norm crystallization has raised two issues addressed in our study. First, how might crystallization best be measured? Traditional measures of crystallization include the standard deviation, the coefficient of variation and the interquartile range (Manning, 1999; Roggenbuck et al., 1991; Shelby & Vaske, 1991). Unfortunately, all of these measures have problems. A newer measure of crystallization, the Potential for Conflict Index (Manfredo, Vaske & Teel, 2003; Vaske et al., 2006), has recently been proposed and is addressed later in this article.

TABLE 1 Hypothetical Frequency Distributions and Conventional Measures of Norm Crystallization

Response Scale Value	Complete disagreement	Uniform distribution	Complete agreement	Skewed agreement
1	50	20	100	10
2	0	20	0	0
3	0	20	0	0
4	0	20	0	0
5	50	20	0	90
Standard deviation	2.000	1.414	.000	1.200
Coefficient of variation	.667	.471	.000	.261
Interquartile range	4.000	2.000	.000	.000

Consider the four hypothetical distributions of a five-value response scale displayed in Table 1. These distributions illustrate important canonical cases, and such ideal-typical situations allow for examination of the face validity of alternative measures of crystallization. The three traditional measures of crystallization do a poor job of distinguishing among these cases. For example, the standard deviation suggests that the skewed agreement case is rather close to the uniform distribution case. It can be logically argued, however, that the skewed agreement case is more similar to that of complete agreement. The standard deviation also yields a value for the uniform distribution case that is closer to the complete disagreement case than to the complete agreement case. However, it can be argued that the value for the uniform distribution case should be approximately midway between the two. The coefficient of variation suffers from the same flaws as the standard deviation, while the interquartile range does not distinguish between the complete agreement and skewed agreement cases. Finally, none of these measures has an upper bound, which makes comparisons between scales of different lengths difficult. Thus, we conclude that a new measure of norm crystallization is needed.

The second issue addressed in our study is what influences norm crystallization as applied in parks and outdoor recreation. Although many studies of normative standards in parks and outdoor recreation have reported some measures of crystallization, little effort has been made to study this issue in a systematic way. However, review of the literature along with consideration of the potential effects of several conceptual and methodological issues gives rise to a series of hypotheses that can be tested using existing data.

Response scale format is an issue of importance in survey research in general but has received only limited attention in normative research applied to outdoor recreation. Hall, Shelby and Rolloff (1996) found that adding a response option allowing respondents to report that “the impact matters to me, but I can't give a number” resulted in higher crystallization of norms for two of the three indicators included in the study. A related issue concerns the number of values included in the response scale. Normative studies use response scales that vary widely with five and nine-point scales used commonly, which may moderate crystallization. A greater number of response options means that a greater opportunity exits to choose a response that differs from someone else's, which leads to the first hypothesis:

• H1. The greater the number of response scale values, the less crystallization will be found.

Normative research in outdoor recreation also uses a variety of evaluative dimensions leading to substantially different normative standards (Manning et al., 1999). This issue may also affect crystallization. For example, the evaluative dimension of “preference” is often used and may result in relatively high levels of crystallization because preference represents the best possible conditions absent all other considerations. There may be less likelihood of agreement on the evaluative dimension of “acceptability” as this notion represents a more marginal and relative standard of evaluation. There may be the lowest likelihood of agreement about “management action” and “displacement”-based standards as these involve judgments about when management actions are needed to protect park conditions or when respondents will choose not to return to a park because of deteriorating conditions. This idea leads to another hypothesis:

• H2. Crystallization will drop as the evaluative dimension used to measure norms moves from “preference” to “acceptability” to “management action”/“displacement.”

The most commonly reported effect on crystallization is the type of area studied, which is most often characterized as backcountry versus frontcountry, defined by levels of use and development. Researchers have found higher levels of crystallization in backcountry areas (Kim & Shelby, 1998; Martinson & Shelby, 1992; Needham, Rollins & Wood, 2004; Shelby, 1981; Shelby & Vaske, 1991; Shelby, Vaske & Harris, 1988; Whittaker and Shelby, 1988) leading to a third hypothesis:

• H3. Crystallization of normative standards will be higher for backcountry than frontcountry areas.

Normative standards have traditionally been measured using narrative and/or numerical descriptions of a range of park and outdoor recreation conditions. However, visual simulations of conditions have been used increasingly (Hall & Roggenbuck, 2002; Kim & Shelby, 2005; Manning et al., 1996; Manning & Freimund, 2004; Needham et al., 2004), and audio simulations have also been used (Pilcher, Newman & Manning, 2008). Given greater information that can be presented in an unambiguous way, these simulations may have “greater face validity” than narrative/numerical approaches (Hall & Roggenbuck, 2002, p. 327). Another hypothesis is:

• H4. Questions that employ visual or audio simulations to describe a range of park and outdoor recreation conditions will generate more highly crystallized norms than questions that use narrative and/or numerical approaches.

As noted earlier, normative studies have addressed a range of impacts and conditions in parks and outdoor recreation that can be broadly categorized as environmental (e.g., trail erosion), experiential (e.g., crowding), and managerial (e.g., acceptability of rules and regulations). Little is known about the relative degree of crystallization of normative standards for these broad categories of indicator variables, though standards for the environmental condition of litter are generally thought to be highly crystallized (Hall et al., 1996). Given the seemingly “objective” nature of environmentally related indicators, crystallization may be higher for these applications than for more subjective experiential and managerial indicators leading to the following hypothesis:

• H5. Normative questions that address environmental impacts/conditions will generate higher levels of crystallization than experiential and managerial impacts/conditions.

Most normative research in parks and outdoor recreation presents a range of park conditions for respondents to evaluate. Ranges include conditions that are favorable to those that are unfavorable. Researchers have not addressed relative agreement across this range of conditions, but greater agreement or crystallization at the extremes of this range (where conditions are highly favorable or unfavorable) might be expected than for more ambiguous conditions in the middle of the range. This reasoning leads to the following hypothesis:

• H6. When respondents evaluate a range of park and recreation conditions (i.e., from low to high levels of impact) there will be more agreement about the extremes of this range than the conditions represented in the middle of the range.

Normative research generally uses two basic question formats that might be described as “long” and “short” (Manning et al., 1999). These long approaches have also been termed “overall evaluation method,” “repeated item,” and “closed-ended.” Short approaches have also been called “specific evaluation method,” “single-item,” and “open-ended” (Kim & Shelby, 2005). Long-form questions ask respondents to evaluate each in a range of conditions, while short-form questions ask respondents to report on conditions that best meet the evaluative dimension specified in the question (e.g., the condition that is “preferred”). Since long-form questions require a series of judgments by respondents as opposed to one judgment in short-form questions, the final hypothesis is presented:

• H7. Respondents who are asked the long-form version of normative questions will exhibit less agreement than those who are asked the short-form version.

Study Methods

A review of the scientific and professional literature in several social science disciplines was conducted to find a suitable measure of crystallization. Based on this review, Van der Eijk's (2001) measure of agreement (A) for ordered rating scales was found to address the shortcomings of conventional measures of crystallization noted above. The formula is:

U denotes the extent of “unimodality” in the distribution, S the number of responses that have nonzero frequencies and K the total number of possible responses. The resulting A statistic is always between −1 and +1, with −1 indicating complete disagreement, 1 complete agreement and 0 uniformly distributed responses across all the values of the scale. Details concerning the computation of U are in Van der Eijk's (2001) paper, but, most basically, it gauges the distance between clusters of responses on the scale. Larger distances indicate less agreement and result in a lower value for U. Its maximum value is 1 in the case of “unimodality” (i.e., no separation between clusters of responses), and its minimum value is −2 in the case of complete disagreement (i.e., maximum distance between clusters) in a three-category scale. The calculation in brackets operationalizes the notion that agreement (A) is larger when the number of nonempty categories (S) is small in relation to the total number of categories (K). Since this calculation must be positive, it follows that U must be negative for A to be negative.

Van der Eijk (2001) begins by considering three simple distributions he calls “unimodal,” in which the value of U is 1 and A is thus computed using only the part of the formula in brackets: (a) all responses are in a single category, (b) the responses are evenly distributed among all the categories and (c) the responses are evenly distributed among two or more contiguous categories, with the other categories having no responses. The first situation denotes complete agreement (A = 1), the second involves a uniform distribution (A = 0) and the third yields one specific type of a general category of distributions Van der Eijk calls semi-uniform. This latter type will yield some intermediate value of A between 0 and 1, depending on the number of nonempty categories in relation to the total number of categories.

To deal with other (and more common) situations, Van der Eijk introduced the notion of “multimodal” distributions. He explained how to decompose them into a series of semi-uniform layers where each layer consists of a mix of categories with zero and nonzero frequencies. As in the type of semi-uniform distribution mentioned above, all nonzero frequencies in each layer are identical, but these frequencies do not necessarily fall in contiguous categories. In this situation the measure of agreement is first computed for each of the layers, and then an overall measure of agreement (A) can be found by weighting the measure for each layer (A_i) by the number of cases in that layer (w _i ) and summing the weighted values:

Complete disagreement is, therefore, a type of “mutimodal” distribution. The skewed distribution from Table 1 is another example of a “multimodal” distribution and, incidentally, yields the value A = .60. This value makes sense (i.e., has face validity) in that is it closer to complete agreement (A = 1) than to the uniform distribution (A = 0) but at the same time far enough from complete agreement to reflect the 10% of the distribution that has a different response scale value than the other 90%.

A related measure, the Potential for Conflict Index (PCI), has recently been proposed to measure crystallization (Manfredo et al., 2003; Vaske et al., 2006). This statistic bears some resemblance to Van der Eijk's measure in that it always falls on a standard scale (0 to 1), with 0 indicating complete agreement and 1 denoting complete disagreement. Thus, it is an improvement over measures such as the standard deviation. However, we chose to use Van der Eijk's measure for two principal reasons. First, the uniform distribution of response scale values does not always yield the midpoint of the PCI statistic, which here would be .5, but rather varies with the number of possible responses or values in the scale. For example, five-, seven- and nine-point response scales give the following values of the PCI for a uniform distribution: .60, .57 and .50. Alternatively, Van der Eijk's A yields a value of 0 (the midpoint of the theoretical range of −1 to 1) in all of these cases. Second, the PCI can be computed only for scales with a neutral center point and an equal number of response options on either side of it (Vaske et al., 2006). The PCI cannot be used for scales with an even number of responses such as strongly agree, agree, disagree and strongly disagree.

To investigate the factors that affect the degree of norm crystallization, a comparative analysis approach was taken (Vaske & Manning, 2008). Data on normative standards of quality collected from surveys in 18 national parks over the course of the past 10 years were assembled for this purpose yielding 1,151 different questions with the number of responses varying from 4 to 25. These studies included a range of contexts (e.g., frontcountry, backcountry), a variety of indicators (e.g., environmental, experiential, managerial), and alternative approaches to measuring normative standards (e.g., different evaluative dimensions and question formats; Manning, Valliere & Jacobi, 1997). Van der Eijk's A was computed for each of the questions to measure crystallization. Ordinary Least Squares Regression was then used to investigate the effects of context and measurement approach on crystallization.

Thus, the dependent variable in the regressions was the measure of agreement applied to the standard of quality indicator. The independent variables included the number of categories for the indicator (i.e., a quantitative variable) as well as the following dummy variables:

• evaluative dimension—preference, managerial action necessary, and displacement versus the excluded category of acceptability;

• context—frontcountry versus the excluded category of backcountry;

• type of presentation—audio/visual versus the excluded category of written;

• type of indicator—environmental impact and managerial action appropriate versus the excluded category of experiential; and

• question format—short form of the questionnaire versus the excluded category of long form.

FIGURE 2 Distribution of Van der Eijk's measure of agreement (A) for normative standards of quality.

Findings

Figure 2 presents a visual display of the distribution of the measure of agreement for the 1,151 questions included in the study. There is a strong propensity for agreement as opposed to disagreement (A = .53 on average, with a standard deviation of .25). Some questions feature complete agreement, but none exhibit the most extreme disagreement, and few are in the negative range at all. About one-third of the questions feature agreement levels between .43 and .66.

TABLE 2 OLS Regression of Measure of Agreement on Possible Determinants of Degree of Crystallization

Determinants of degree of crystallization	Unstandardized regression coefficient	Standard error	Standardized regression coefficient	T	P-value
(Constant)	.473	.071		6.628	.000
Number of categories	−.010	.005	−.076	−2.127	.034
Preference	.086	.028	.093	3.012	.003
Managerial action necessary	−.062	.028	−.068	−2.220	.027
Displacement	−.001	.028	−.001	−.035	.972
Frontcountry	−.035	.027	−.041	−1.273	.203
Audio/visual presentation	.238	.019	.353	12.438	.000
Environmental impact	.051	.020	.075	2.565	.010
Managerial action appropriate	−.116	.052	−.071	−2.238	.025
Note. R-squared = .181.

Table 2 provides the results from an Ordinary Least Squares Regression of the measure of agreement on the possible determinants of the level of agreement. As suggested in Hypothesis 1, the number of categories has a negative effect on crystallization, though the size of this impact is small. When the number of categories increases by 1, the measure of agreement on average declines .01. Thus, Hypothesis 1 is confirmed.

Hypothesis 2 is partially confirmed in that what is preferred exhibits more agreement than what is minimally acceptable. The situations that respondents feel call for management action feature less crystallization than those found to be minimally acceptable. However, contrary to the hypothesis, the displacement scenario displayed no difference with the acceptable condition and had a slightly higher level of crystallization than the management action situation (test not shown in table, though T = 1.836, p = .067).

Somewhat surprisingly, and contrary to Hypothesis 3, no difference was found in crystallization between visitors to frontcountry and backcountry parks (or these respective portions of parks). This issue is addressed more fully in the discussion section. A large difference in crystallization between questions that used visual/audio simulations of impacts and those that did not confirmed Hypothesis 4. As predicted by the fifth hypothesis, the results indicated a higher level of agreement when questions concerned potential environmental impacts than when they queried more experiential issues such as crowding and managerial issues (e.g., acceptability of alternative management actions).

Hypothesis 6 was tested in a somewhat different manner than the others. Respondents evaluated each condition in a range of conditions only in the “long-form” version of the normative question. From the 1,151 study questions, 826 were from the long-form version. In this version nearly all the questions employed the evaluative dimension of acceptability, so replicating the model in Table 2 was not possible. Instead, level of agreement was compared among the three different order placements: the beginning third, the middle third and final third of the range. As predicted, low and high order conditions featured more agreement than those in the middle. Specifically, the mean level of agreement for the conditions at the beginning was .63, compared to .41 for those in the middle and .48 for those at the end. All the differences among these were statistically significant.

TABLE 3 OLS Regression of Measure of Agreement on Possible Determinants of Degree of Crystallization: Short Form Versus Long Form

Determinants of degree of crystallization	Unstandardized regression coefficient	Standard error	Standardized regression coefficient	T	P-value
(Constant)	.331	.058		5.749	.000
Frontcountry	−.030	.028	−.036	−1.103	.270
Audio/visual presentation	.238	.019	.353	12.329	.000
Environmental impact	.057	.020	.083	2.876	.004
Managerial action appropriate	−.103	.052	−.063	−1.980	.048
Short form	.043	.016	.076	2.706	.007
Note. R-squared = .165.

The final study hypothesis was examined using the results in Table 3. Some of the variables from Table 2 are not present in Table 3 (i.e., the number of response scale categories and those having to do with the evaluative dimension such as preference, acceptability, displacement and managerial action). This result occurred because the correlation between questionnaire version and the omitted variables was near perfect. The long-form version of the questionnaire employed 9-category scales in 99.2% of the items, and 99.4% of the time the evaluative dimension queried was acceptability. This table seems to provide some support for Hypothesis 7. Crystallization was somewhat higher on the short form as opposed to the long form (.043). However, in Table 2 the effect of the number of categories is about −.01. Since the long-form version of the questionnaire had about three more categories on average than the short-form version (i.e., 9 versus 6), this effect is believed to be largely an artifact of the number of categories. The other variables in this table display effects similar to what was found when they were examined in connection with Table 2.

Discussion

In this study we proposed and applied a new measure of norm crystallization. This measure appears to represent an improvement over conventional measures. Use of this new measure can allow for more informative descriptive and analytical research on the extent of normative agreement among visitors to parks and related areas.

Descriptive results from this study indicated a fairly high level of overall norm crystallization: .53 on average on a −1 to +1 scale. In addition, there were almost no instances where crystallization was less than 0. This finding suggests that normative research can often contribute in important ways to formulation of standards of quality for parks and outdoor recreation and natural resources management more broadly, because of generally a relatively high level of agreement (crystallization) among visitors and other stakeholders about appropriate conditions.

TABLE 4 Responses to Questions about Standards of Quality that Exhibit an Average Level of Crystallization

Study site: Schoodic Peninsula at Acadia National Park		Study site: Haleakala National Park Visitor Center
Item: Which photograph shows the number of people that would be so unacceptable that you would no longer visit Schoodic Point?		Item: Which photograph shows the number of people you would prefer to see?
Photo	Percentage of respondents	Photo	Percentage of respondents
1 (0 PAOT)	.67%	1 (0 PAOT)	18.57%
2 (33 PAOT)	3.78%	2 (15 PAOT)	45.71%
3 (66 PAOT)	19.33%	3 (30 PAOT)	25.71%
4 (99 PAOT)	38.22%	4 (45 PAOT)	5.00%
5 (132 PAOT)	38.00%	5 (60 PAOT)	5.00%
Note. PAOT = People at One Time.

More specifically, though, what does this generally “high” level of agreement mean? To develop insights about this question, Table 4 shows frequency distributions for two different five-category questions, each of which features an average level of crystallization found in the comparative analysis (Van der Eijk's A in the range .50 to .55). In each case, more than 70% of respondents chose one of two contiguous photographs representing the level of use that would cause these respondents to be displaced from the study areas. A substantial level of agreement among visitors about appropriate conditions in parks and related areas seemed evident. Such data can help guide formulation of standards of quality in empirical and definitive ways.

The analysis of the determinants of the level of norm crystallization yielded interesting findings as well from methodological and substantive points of view. Regarding the former, visual and audio simulations elicited significantly higher levels of agreement than narrative/numerical approaches. Since natural resource experiences are often aesthetic in character, using visual approaches for research is important where appropriate. Otherwise, less norm crystallization than is actually present might be reported.

The number of categories included in response scales is not highly associated with the level of agreement. Net of other factors, a five-point scale will have a level of agreement that is only about .04 higher than a nine-point scale. The primary reason to consider limiting the number of categories in response scales is probably to reduce respondent burden.

Regarding issues of substance, what respondents deemed preferable displayed more crystallization than what they found minimally acceptable, and their opinions on environmental impacts featured more agreement than their evaluations of experiential conditions. In contrast, respondents' views concerning when management action would be necessary were less consistent than their judgments regarding acceptability. Similarly, views on the appropriateness of different types and levels of managerial actions featured less crystallization than evaluations of experiential conditions. Taken together, however, these substantive effects were small. R-squared increases only by .015 (from .142 to .167) when the substantive variables are added to the methodological variables. This, too, suggested the potential usefulness of normative data. Irrespective of the type of question, the level of agreement regarding norms is generally high.

The most surprising substantive finding was the lack of difference in crystallization between normative standards for backcountry and frontcountry areas. Several prior studies have found such differences and all suggesting more highly crystallized norms in backcountry areas (Kim & Shelby, 1998; Martinson & Shelby, 1992; Needham et al., 2004; Shelby, 1981; Shelby & Vaske, 1991; Shelby et al., 1988; Whittaker & Shelby, 1988). The contrary findings of this study might be at least partially explained by another structural property of norms called norm “salience” (see Figure 1). Salience refers to the importance of the indicator (i.e., impact/condition) to respondents. The more salient an indicator, the more highly an associated norm might be crystallized (because it has been given more thought and attention). Normative research in outdoor recreation was initially applied to backcountry contexts and addressed the most salient indicators in these locations such as number of other visitors encountered. However, crowding and other recreation-related impacts may be manifested differently in frontcountry contexts. For example, problems associated with high use levels in frontcountry areas may be more directly and more importantly manifested as waiting times for essential services or opportunities than in the number of other visitors encountered. Yet, normative research in frontcountry areas, especially conducted in initial studies, may be addressing conventionally backcountry-oriented indicators—indicators that are not highly salient in frontcountry—and may be why crystallization is relatively low. Many of the studies included in the comparative analysis presented in this article are more recent and employ indicators that may be more relevant in frontcountry areas. For example, the study of Statue of Liberty National Monument addressed waiting times to climb the statue and board ferries rather than the number of visitors encountered.

Conclusion

Van der Eijk's measure of agreement (A) offers an improved statistic to assess crystallization of norms in parks and outdoor recreation. This measure can be interpreted intuitively and does not suffer from the inconsistencies of more conventional measures, including the standard deviation, coefficient of variation, and interquartile range.

Application of this statistic to 1,151 normative questions applied in 18 national parks found a generally high level of crystallization (i.e., A = .53 on average with a standard deviation of .25). A comparative analysis using these data found that crystallization can be affected by several substantive and methodological variables including:

• number of categories in response scales,

• evaluative dimension employed,

• use of visual/audio simulations versus narrative/numerical descriptions of impacts/conditions,

• type of impact addressed,

• low versus medium versus high levels of impact, and

• long versus short form of normative questions.

However, these effects are relatively small, suggesting that norm measurement is relatively robust, at least with respect to crystallization or level of agreement. The most important independent variable in this study was the use of visual/audio simulations to represent the range of impacts/conditions under study. This approach should be used where indicator variables lend themselves to this treatment.

We recommend that Van der Eijk's measure of agreement be adopted in normative studies to assess the degree of consensus about acceptable conditions in parks and outdoor recreation or the extent to which there are social norms about this issue. Moreover, the relatively high levels of crystallization found in the comparative analysis, along with the general robustness of normative questions, suggest normative data can be useful in formulating standards of quality in parks and outdoor recreation and natural resources management more broadly.