Exploring the dimensions of place: a confirmatory factor analysis of data from three ecoregional sites

Abstract

Themes of place, situatedness, and locale are increasingly prominent in environmental education literature and practice. Sense-of-place research, which considers how people connect with places and the influence of those connections on engagement with the environment, may have important implications for environmental education. Prior place studies have proposed that people’s place connections have various dimensions. This paper explores four place dimensions, analyzing data from a survey (n = 712) conducted in three ecoregional sites in which we investigated residents’ sense of place. We examine how our data fit a proposed typology of place dimensions – a four-dimension (biophysical, psychological, sociocultural, and political-economic) categorization based on previous conceptions of the dimensions of place. We use structural equation modeling to question whether a 4-factor conceptualization of the dimensions of sense of place is superior to plausible alternatives. In comparing the four-dimension model with alternative models, we find that the four-dimension model is the best fit for these data. Our findings suggest that environmental and place-based education may benefit from an understanding and consideration of this four-dimension conceptualization of place in program design, implementation, and evaluation.

Keywords:

• sense of place
• place-based education
• typology of place dimensions
• structural equation modeling

Acknowledgments

The lead author is grateful to Professors Stephen Kellert, William Burch, and Carol Carpenter at Yale University’s School of Forestry and Environmental Studies for their guidance on the overall research on which this article was based. We appreciate the editor’s and anonymous reviewers’ constructive comments, which strengthened this article. Finally, we appreciate the generous support provided for this study by WWF-US, the Charles Darwin Foundation, the Mountaineers Foundation, Project AWARE, the Switzer Foundation, the Tropical Resources Institute, Morris K. Udall Foundation, the Urban Resources Initiative, the Yale Institute for Biospheric Studies, and the Yale Center for International and Area Studies.

Notes

aAsymptotic covariance matrix available upon request.

bPolychoric correlation matrix for the primary sample is presented in the upper right triangle.

cPolychoric correlation matrix for the holdout sample is presented in the lower left triangle.

1. Landscape historian Jackson (1994, 57) laments that, ‘“sense of place” is a much used expression, chiefly by architects but taken over by urban planners and interior decorators and the promoters of condominiums, so that now it means very little.’ Stedman (2003, 824) agrees that, ‘sense of place research is fraught with nagging questions about concept and clarity.’

2. An ecoregion is ‘a relatively large unit of land or water that contains a distinct assemblage of natural communities, sharing a large majority of species, dynamics, and environmental conditions’ (Dinerstein et al. 2000, 15). Ecoregions are delineated into those that are representative of freshwater, marine, and terrestrial systems (Olson and Dinerstein 1998, 3–5).

3. The items and associated findings are discussed further in Ardoin (2009).

4. Internal consistency for each set of measures was assessed according to the guidelines recommended by DeVellis (2003).

5. We drew on past work in community and place attachment (e.g. Hummon 1992; Low and Altman 1992; Williams and Vaske 2003) to conceptualize framing and items appropriate to address social relationships with place. We then created items specifically for this study to ensure fit with the larger context of our survey.

6. Because these cases all consisted of repeating patterns of 4s or 5s, removing them also helped to lower skewness and kurtosis within the data. Considering the possibility that these 31 cases represented legitimate response patterns and that their consistency supported the single-factor plausible alternative to the proposed 4-factor oblique model, we ran the analysis both with and without these cases. The single-factor model was consistently inferior.

7. Although the general rule of thumb for inclusion in CFA is often skewness and kurtosis values of ±1, Schumacker and Lomax (2004) indicate normal theory can be applied when skewness and kurtosis are within the range of ±1.5.

8. Kelloway (1998, 22) notes that holdout samples are a means of replicating post hoc modification to a model based on a primary sample.

9. Because WLS can be unreliable with a sample size of less than 2000, the results of the CFA based on WLS estimation were checked using a second method, maximum likelihood estimation (ML) with the Satorra–Bentler scaled chi-square (also based on polychoric correlation and asymptotic covariances matrices). When ordinal data are treated as interval-level data, as can be done with Likert-scale variables, and estimated with ML, the Satorra–Bentler chi-square is recommended to help correct model chi-square for multivariate nonnormality (Joreskog 2004). According to Schumacker and Lomax (2004), this approach is acceptable when ordinal variables have at least five categories, and their skewness and kurtosis values are within the range of ±1.5 (63). Throughout, the fit statistics generated from the ML-estimated CFAs were equivalent with or superior to those generated from WLS-estimation of the model, such that the initial and revised 4-factor models (using both the primary and holdout samples) met the criteria for a good fit according to the four key fit indices: NNFI (.97 to .99), CFI (.98 to .99), RMSEA (.046 to .058), and SRMR (.049 to .08; more data available on request). Given that WLS is both the more appropriate method of estimation and yields more conservative results, WLS are reported as the primary statistics in this article.

10. Although there was a slight negative correlation between S4 and P6 (−.04), both items were retained because the value approached zero and was reasonable given the proposed factor structure. We ultimately removed S4 from the revised model based on modification indices.

11. When the CFA was run including P4, the phi was not positive definite, affirming the decision to remove the variable.

12. In SEM, a nonsignificant χ ² is ideal, indicating that the model is an excellent fit for the data; however, χ ² is sensitive to sample size: with moderate to large samples, χ ² tends to be overly conservative with increased type II error (i.e. increased chance of falsely rejecting a true fit) and is thus not recommended as a reliable fit statistic when n > 200 (Garson 2011).