Is cultural heritage really important for tourists? A contingent rating study

Abstract

In this article we present the results of a contingent rating study carried out on a sample of tourisits visiting Scicli, a Sicilian town known for its baroque heritage. In particular, we focus on different attributes of tourism products – namely, season, accommodation and cultural heritage – to study how much each of these attributes weights in tourists’ preferences. We also study how the socio-demographic characteristics of people affect their evaluation of the different attributes of tourism products. The heritage endowment appears to be far from being the most important factor; this result is consistent across different socio-demographic subgroups of interviewed persons.

Acknowledgements

We are indebted to Rosa Pacetto who has collected the data through direct interviews. We also thank Massimiliano Mazzanti, Antonello E. Scorcu, Giovanni Signorello, and an anonymous referee for helpful comments. The usual disclaimer applies.

Notes

¹ The interviewer was Ms Rosa Pacetto, and the interview project was part of her final dissertation for the Laurea Degree at the University of Catania. The database containing all answers is available from Authors on request, in Excell, Microfit or Limdep format.

² It is important to stress that the estimation of the willingness to pay is not among our goals. The declarations on WTP will be used only for a different purpose: we will limit ourselves to analyse how the declared willingness interacts with elicited preference system.

³ Two further reasons exist to omit price from the list of attributes: first, we do not aim at estimating implicit price of attributes, so that we do not need the presence of price among the explanatory factors; second, in similar exercises, price typically emerges to have a positive marginal coefficient in the evaluation system of a package - see, e.g. Roe et al . (1996) or Alberini (2003) among many others; the reason is that respondents interpret price as an indirect indicator for quality. The absence of price avoids this source of confusion.

⁴ For comprehensive reviews, see Cattin and Wittink (1982), Carrol and Green (1995), Green et al . (1985), Green and Krieger (1997); more recently, Hanley et al . (2001), Cuccia and Signorello (2002), Cuccia (2003) and Mazzanti (2003). Montecarlo evidence on conjoint analysis are provided by Carmone et al . (1978).

⁵ See Alvarez-Farizo et al . (2001) or Alberini et al . (2003) as examples of recent conjoint analyses in which the personal attributes are inserted into the general regression.

⁶ Conjoint analysis and more specifically contingent rating represent indirect stated preference methods to draw the evaluation of people on the attributes of goods; differently, contingent valuation method represents a direct stated preference method which basically consists of asking individuals to declare their personal valuation of a nonpriced asset (Mitchell and Carson, 1990; see also Arrow et al ., 1993, and Diamond and Hausman, 1996); recent examples of contingent valuation study are Pavlova et al . (2004), or Barreiro et al . (2005); in the latter a particular focus is set on the individual characteristics of respondents.

⁷ In order to have a comprehensive view about the problems concerning the sample selection and the description of products, see, e.g. Green and Srinivasan (1978, 1990), Cattin and Wittink (1982), Green et al . (1985), Carrol and Green (1995), Hanley et al . (2001), Cuccia (2003), and the articles published in the recent issue of the Journal of Cultural Economics (2003) on contingent valuation.

⁸ See, e.g. Hanemann (1984), McConnell (1990) and especially Roe et al . (1996) who transform the grades (analysed through a ordered-logit estimation) into a dichotomous variable (analysed through a logit estimation) and compare the different results.

⁹ The difference between logit and probit estimations rests on the assumption about the error distribution. With the polychotomous dependent variable, we performed also the ordered-logit estimation, obtaining – as it is usual – very similar results as compared with the results from ordered-probit. Also in the case of transformed dichotomous variable, the results from logit and probit estimations are very similar.

¹⁰ Our findings are consistent with the evidence presented by Mackenzie (1993), who empirically compares three different response formats (rating, rankings and binary choice) and shows that ratings provide – as expected – the largest informational efficiency in econometric estimation.

¹¹ We have also investigated the relevance of the individual demographic variables upon the marginal effect of attribute levels, by inserting, in the regression, explanatory variables capturing the cross-combination of attribute levels and demographic characteristics (see, e.g. Begona et al ., 2001, and Alberini et al, 2003, among many others); such combined explanatory variables, however, are not significant once we consider demographic variables per se and attribute levels, so that we omit the results for the sake of brevity. For a recent study concerning the role of socio-economic variables in the demand for cultural goods, see Borgonovi (2004).

¹² From a methodological perspective, note that the number of the proposed levels for each considered attribute should have no effect on the weight attached to the attribute. In fact, the present results –like the available literature– show that such effect does not exist indeed. In other words, the fact that four level of accommodation are included in the prospects offered to respondents compared to only two each for season and cultural visits, in the present case, should not affect the results concerning the weight.

¹³ This result can depend on the larger income declared by the sub-group of men as compared to the whole sample.

¹⁴ In front of these marked differences, it is not surprising that the Chow test of stability of the regression coefficients – with respect to the OLS estimation – leads to reject the stability of coefficients computed for the female sub-groups over the next observations regarding the male group (, p = 0.000).

¹⁵ These results resemble the evidence concerning the distinction between female and male respondents. However, it is important to notice that the gender distribution in young and old subsamples reflects the distribution in the whole sample: the men are 42.34% of the whole sample, the 42.30% and the 42.37% of the young and old respondents, respectively. In other words, the fact that old people show a positive attitude toward high level accommodation is not due to the fact that in this sub-sample there is a larger presence of men.

¹⁶ Not surprisingly in front of the regressions results, the Chow test of stability of the regression coefficients – with respect to the OLS estimation – leads to accept the stability of coefficients computed for low-educated people over the next observations regarding highly educated persons (, p = 0.94).