Revisiting the Listening Styles Profile (LSP-16): A Confirmatory Factor Analytic Approach to Scale Validation and Reliability Estimation

Abstract

The Listening Styles Profile (LSP-16) was developed to measure an individual's preferred listening style. One frequent criticism of the LSP-16 is the consistently low estimates of internal consistency. The following study addresses this concern using confirmatory factor analysis to assess both the latent constructs of the scale (i.e., People, Content, Action, Time) and the scale's reliability. Results suggest that listening style is multidimensional; however, additional scale development is needed to increase subscale reliability estimates. Suggestions for future research and development are provided.

A previous version of this manuscript was presented at the 2007 International Listening Association Convention.

Notes

¹This makes sense when considering the purpose of the chi-square statistic: to compare an observed matrix to an expected matrix. A nonsignificant chi-square, therefore, suggests that the observed matrix is the same as the expected (i.e., theoretical) matrix. In other words, the sample data conform to the proposed factor structure.

²For those familiar with exploratory factor analytic methods, the residual matrix can be reproduced within those programs as well. In fact, within the framework of PCA, the model that converges has the lowest possible values in the residual matrix (which is why some factor analyses take more iterations to converge).

³In the language of regression, a residual is, for any given data point, the difference between the value of the dependent variable and the predicted value of the dependent variable as estimated from one or more independent variables. As in regression where the dependent variable is never perfectly predicted by the set of independent variables, in CFA the theoretical model is rarely perfectly predicted by the sample data.

⁴Basically, if all values equal zero then the sample data exactly replicated the theoretical model and chi-square would also equal zero. When chi-square is not zero (and is significantly different from zero), we can expect the residual covariances to deviate from zero as well.

⁵All analyses are available from the first author upon request.

⁶Maximum likelihood method uses a fitting function analogous to the least squares criterion in regression.

⁷Indeed, removal of the content factor, all four content items, and the item error terms produced a well-fitting model, χ² (51) = 124.91, p < .001, fit ratio = 2.45, CFI = .93, RMSEA = .05 (CI90% = .04, .06).

⁸For all alternative (and nested) models, the way in which we determined one model as “better” than another was by comparing the two chi-square values. Two models can be compared by looking at a chi-square change score that is generated by subtracting the smaller of the two chi-square values from the larger (Kline, 2005). The statistical significance of this value is determined by comparing the change value to a critical value found in a chi-square table (Tabachnick & Fidell, 2007).

⁹As one reviewer suggested, although the “four-factor model most accurately captures listening styles among the models examined [it] could be that a model with more than four factors would be a better fit for the data (even though there was no theoretical reason to test for such a model in this case).” To test this assertion, we used several data driven methods. None of the exploratory factor analyses suggested the extraction of more than four factors. In fact, all exploratory analyses suggested the interpretation of four factors. All analyses are available from the first author upon request.

¹⁰The formula used for the point estimation of scale reliability was

where (Σλ _i )² is the squared sum of unstandardized regression weights and Σθ _ii is the sum of unstandardized measurement error variances.