Investigating Visitor Profiles as a Valuable Addition to Museum Research

Abstract

There is a long tradition of museum research assessing visitors' personal background. In this article, we suggest an insightful way to enhance and intensify visitor analyses and adopt a more integrative approach. To this end, we draw attention to Latent Class Analysis (LCA), a classification method that allows us to investigate visitor profiles rather than isolated characteristics while using a mixture of socio-demographic variables and other personal characteristics. We illustrate our suggestion with two examples drawn from a study we conducted at two science museums and two museums of cultural history. The first example shows the use of LCA on variables that are easily assessed and that many museums already have access to through their visitor surveys. The second LCA involves more specific cognitive and motivational visitor variables that are of general importance for learning and that are assumed to be especially important when dealing with conflicting information. Both examples illustrate the additional value of investigating visitor profiles for researchers as well as museum practitioners.

Keywords:

• Museum
• Visitor research
• Latent class analysis

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. In order to decide which of the LCA solution fits best we looked at Akaike Information Criterium (AIC) and Bayesian Information Criterium (BIC) as measures for model comparison and the Lo-Mendell-Robin Likelihood Ration Test (LMR-LRT) to show whether there is an improvement in fit between neighboring class models (i.e. comparing k–1 and k class models) (Nylund, Asparouhov, & Muthén, 2007). In addition, entropy as a measure of classification certainty is reported. Fit values for two classes: AIC = 6,204.39; BIC = 6,258.40; Entropy = .69; classification probabilities: class 1 = 0.98, class 2 = 0.88; Lo–Mendell–Rubin LRT = 292.80; p < .001. Fit values for three classes: AIC = 6,196.44; BIC = 6,279.91; Entropy = .85; classification probabilities: class 1 = 0.95, class 2 = 0.99, class 3 = 0.52. Lo–Mendell–Rubin LRT = 19.48; p < .01. Fit values for four classes: AIC = 6,197.84; BIC = 6,310.71; Entropy = .80; Entropy: classification probabilities: class 1 = 0.91, class 2 = 0.90, class 3 = 0.72, class 4 = .98. Lo–Mendell–Rubin LRT = 10.39; ns.

2. Fit values for two classes: AIC = 14,339.63; BIC = 14,447.56; Entropy = .77; classification probabilities: class 1 = 0.92, class 2 = 0.94; Lo–Mendell–Rubin LRT = 703.81; p < .001. Fit values for three classes: AIC = 14,021.40; BIC = 14,168.57; Entropy = .80; classification probabilities: class 1 = .91, class 2 = 0.87, class 3 = 0.93. Lo–Mendell–Rubin LRT = 328.30.39; p < .001. Fit values for four classes: AIC = 13,788.39; BIC = 13,974.81; Entropy = .78; classification probabilities: class 1 = 0.87, class 2 = 0.84, class 3 = .91, class 4 = .89: Lo–Mendell–Rubin LRT = 244.50; ns.

3. The results of this analysis should be interpreted with care, as the estimated class assignment was fixed for comparison of means with SPSS. Therefore, it is possible that there are slight deviations in the results of the fixed class assignments compared with the estimated classes with Mplus. However, the rather high average probabilities in the LCA for most likely latent class membership (> .85) indicate that the deviations will be very low.

Additional information

Funding

This work was supported by the German Research Foundation under Grant LE1303/8-1.