1 Introduction

Cultural amenities have emerged as a potential driver of accumulation of human capital across cities (Falck et al. 2011; Buettner and Janeba 2016; Beckers and Boschman 2019). Skilled workers—who display strong tastes for urban amenities—tend to locate in cities offering greater varieties of urban and leisure amenities (Carlino and Saiz 2019). Consequently, as early suggested by Glaeser et al. (2001), cities are not only centers of production, but also shopping centers, with a growing variety of goods and services, where larger cities stand out because they offer a higher variety of cultural goods (Schiff 2015; Lopez 2019).

While cultural economics and urban economics literatures have well established that schooling is one of the main determinants of cultural consumption (Hallmann et al. 2017), as more able individuals tend to prefer cities with a high level of cultural amenities, there is no empirical evidence on whether cities specialized in culture boost cultural consumption of their residents. This paper fits into this framework by directly testing whether––after controlling for individual characteristics––the higher the city cultural supply, the higher the probability of consuming cultural goods.

To the best of our knowledge, only a few studies have studied the interaction between city characteristics and individual consumption of culture. For instance, Prieto-Rodríguez and Fernández-Blanco (2000) investigate popular and classical music in Spain by assessing the importance of regional variables. More recently, Courty and Zhang (2018) include city dummies and find important differences in cultural participation across cities in China. In this context, this paper makes two contributions to the existing literature. First, it provides direct empirical evidence on whether an increased city cultural supply rises workers’ consumption of cultural goods. Second, it offers valuable information to support local policies aimed at enhancing the attractiveness of cities.

This study focuses on Chile, a developing country with a heterogeneous supply of cultural infrastructure across space and a high concentration of people in the capital city of Santiago, located in the Metropolitan Region.Footnote 1 To perform the empirical analysis, we combine several city-level attributes with data from the National Survey of Cultural Participation (ENPCC, by its initials in Spanish), developed by the Ministry of Culture, Arts and Heritage of Chile in 2017 to assess the impact of cities’ cultural supply on the probability of consuming cultural goods for the Chilean workers.

Using an instrumental variable (IV) approach, based on, partially, the shift-share analysis (Bartik 1991; Partridge and Rickman 2008; Moretti 2013), the main findings suggest that cities’ cultural supply exert a significance influence on workers’ decisions to consume cultural goods, this result is found for aggregate as well as for separated categories of cultural goods. Specifically, doubling the city-level of cultural supply would produce an increase in the estimated marginal effects that an average individual consumes at least five cultural goods by 1%. Further analyses confirm the robustness of the relationship—except for performing arts—between city-level cultural supply and individual consumption of cultural goods, even under a strong degree of violation of the exclusion restriction of the instrumental variable, and the use of multilevel models to account for the spatial aggregation of individuals into cities.

The remainder of this paper is organized as follows. Section 2 presents a brief literature review on cultural consumption. Section 3 outlines the data. Section 4 sketches the empirical strategy and discusses some identification issues. The main results are presented in Sect. 5, and finally Sect. 6 concludes.

2 Cities and cultural consumption

Scholars have usually studied individual’s consumption of cultural goods using the model of leisure demand developed by Humphreys and Ruseski (2011), where a worker first must decide whether to consume culture and then how to allocate his/her time to this and other activities such as the time spent on paid work.Footnote 2

Using this theoretical framework, the literature has mostly devoted to estimate the impact of several individual characteristics, such as available time, income and human capital on the consumption of cultural activities. One of the main findings is the robust and strongly positive effect of human capital (education) on the probability of participating in cultural activities (Muñiz et al. 2011; Prieto-Rodríguez and Fernández-Blanco 2000; Favaro and Frateschi 2007; Hallmann et al. 2017). Similarly, empirical studies identify a positive relationship between income and participation in cultural activities (Ringstad and Løyland 2006, 2011; Favaro and Frateschi 2007; Muñiz et al. 2011; Courty and Zhang 2018).

Regarding the theoretical examination of city-level cultural supply and individuals’ cultural consumption, Rössel and Weingartner (2016) develop a theoretical model that relates actual cultural consumption, cultural infrastructure, individual preferences, and socio-structural position of individuals. These authors assert that both individual and city characteristics influence the consumption of cultural goods, and more importantly, city attributes play a major role as external conditions—beyond individual control—limiting what an individual does and chooses to consume.Footnote 3

In this context, larger cities stand out because they would exhibit a higher concentration of both art and cultural activities, a typical result derived from the advantages of agglomeration externalities produced by the spatial concentration of the economic activity in certain places. Furthermore, such an excessive concentration of firms and people would produce an environment that boosts the development of such creative activities (Diniz and Machado 2011; Markusen and Schrock 2006). Following these ideas, Muñiz et al. (2011) test whether a higher consumption of culture might be expected in larger cities and they find that—although the results were not significant in all cases—participation in cultural activities is positively influenced by the presence of major cultural facilities in big cities in Spain. Similarly, Rössel and Weingartner (2016) test whether the supply of culture in a geographical area affects the actual extent of cultural consumption in Switzerland. The authors conclude that the availability of cultural attractions and infrastructure contributes to explain cultural consumption; however, individual characteristics remain of greater importance to explain the consumption of cultural goods.

3 Data

This research uses the ENPCC, developed by the Ministry of Culture, Arts and Heritage of Chile in 2017. This survey makes a characterization of the cultural participation in Chile, collecting information of population of 15 years of age or older who resides in urban areas with populations ≥ 10,000 inhabitants. It is statistically representative of 135 urban areas with a coverage of 85% of the target population, and for data gathering, a probabilistic sample design was used, geographically stratified by region and size of the census blocks according to their number of houses (INE 2018). The selected sample for this study consisted of 5965 workers composed of men between ages of 15 and 65 and woman aged between 15 and 60 years. After dealing with missing values among the key variables, the final sample corresponds to 3166 observations distributed across 117 cities that represents around 34% of total Chilean cities including the political capital of Santiago.Footnote 4 Since part of the observations is lost during the preparation of the final sample, and although the original sample is representative of the 85% of the total target population, we prefer a cautious interpretation of the results and treat them as representative only of the selected sample.

This survey contains several individual characteristics, but more importantly it includes data regarding the consumption of nine types of cultural goods: theater, modern dance and ballet spectacle, opera, classic music concerts, live music concerts, circus performances, movie theaters, museums, and art displays. To measure cultural consumption, respondents are asked the following question: In the last 12 months, have you ever attended to the following cultural activity? This question is asked for each of the nine cultural categories.

To perform the empirical analysis, we first focused on the aggregate level of cultural consumption by creating an ordered categorical variable between 1 and 5 according to the number of different cultural activities an individual consumes (see Table 1 for details). It is important to consider the following when interpreting the results, this variable captures simultaneously both variety and intensity of the cultural consumption. For example, if an individual consumes three of the nine types of cultural goods included in the survey, this reflects both a variety in the consumption since the individual consume three different types of goods, but also the intensity of his/her cultural consumption as a whole. While about 52% of the sample reports to consume between zero and one cultural good, 7.3% reports to consume five or more (up to nine) cultural goods. Although the aggregate analysis is a valuable empirical task, the literature has extensively documented that cultural consumption can be divided into three groups according to the tastes, preferences and practices of individuals. These groups correspond to highbrow, middlebrow and lowbrow cultural activities and it is expected their consumption patterns differ notably among individuals (Katz-Gerro and Jaeger 2013; Domański 2017). Following this recommendation, we grouped cultural consumption into three categories: (1) performing arts (theater, modern dance and ballet spectacle, opera, classic music concerts, live music concerts, and circus performances); (2) movie theaters; and (3) museums and art displays.

Table 1 Worker and city characteristics: definition and summary statistics

To control for individual characteristics, we included measures of schooling, age, gender, marital status, and the number of people living in the worker’s household.Footnote 5 We also developed specific variables to account for the worker’s available time, such as the weekly hours dedicated to working and commuting, weekly hours dedicated to domestic labor and caring for children and those dedicated to resting and leisure activities, following Hallmann et al. (2017).

Regarding city variables, the most important city attribute corresponds to the cultural supply, however, as stressed by Glaeser (2005) and Markusen (2006), measuring culture on an aggregate scale is challenging. For example, regular measures of cultural employment, as those employed by Throsby (2010) and Florida (2002), which use the broadest occupational categories as a base, face limitations due to the inclusion of major categories of occupations, tending to overestimate the size of cultural employment.

Due to the above-mentioned concerns, this research follows Boualam (2014) to construct a proxy for cultural supply that includes only occupations devote to produce cultural goods and services that are consumed locally. To identify occupations that meet the above-mentioned definition of cultural employment we employed Chile’s 2017 National Socioeconomic Characterization Survey (CASEN). Using this information, city-level cultural supply consists of a measure between 0 and 100, that represents the share of workers in the cultural sector with respect to the total city employment.

The literature has also stressed that temperature and precipitation affect cultural consumption, especially museum attendance, as stressed by Cellini and Cuccia (2019) and Cuffe (2018) for Italy and New Zealand, respectively. In Chile, climate conditions are highly heterogeneous across space, therefore we also control for temperature and precipitation.Footnote 6 The consumption of cultural goods can compete or have a relationship with other leisure activities, such as sports (Muñiz et al. 2011; Hallman et al. 2017) and video game playing (Borowiecki and Prieto-Rodriguez 2015). Due to the above, we also control for the number of historical monuments per capita and the square kilometers of green areas in each city, to have some insights on the association between them and cultural consumption.

To have a measure on city accessibility, the average city commuting time was computed using data from the 2017 CASEN survey. Finally, since tastes for amenities can vary according to the population density (Backman and Nilsson 2018), we include a measure for population density. Table 1 reports the summary statistics and data description.

4 Empirical strategy

As an empirical strategy, we take as a base a random utility model (McFadden 1974), commonly used for the analysis of individuals’ decisions of consumption with a discrete dependent variable. In this model, a worker decides whether to consume or not consuming culture, depending on which option generates more utility. Since our main dependent variable is an ordered categorical variable between 1 and 5 according to the number of cultural activities an individual consumes, we use an ordered probit model (OPM).Footnote 7

Since we do not observe the continuous consumption of cultural goods of workers, we assume that the true latent cultural consumption \({\text{CC}}^{*}\) of worker k located in city c, depends on the years of education \(\left( {{\text{sch}}_{{{\text{kc}}}} } \right)\), city level cultural supply (\(\sup_{c}\)), a vector of control variables (\(x_{{{\text{kc}}}}\)), and a stochastic error term (\(\varepsilon_{{{\text{kc}}}}\)). This latent cultural consumption \({\text{CC}}^{*}\) can be stated as:

$${\text{CC}}_{{{\text{kc}}}}^{*} = \beta {\text{sch}}_{{{\text{kc}}}} + \alpha \sup_{c} + \gamma x_{{{\text{kc}}}}^{^{\prime}} + \varepsilon_{{{\text{kc}}}}$$

where \({\text{CC}}^{*}\) is a latent (unobserved) continuous measure that ranges from \(- \infty\) to \(+ \infty\). We can operationalize this unobserved measure using the observed discrete variable with categories that represent the workers’ cultural consumption. We link this discrete category variable with the continuous latent measure \(\left( {{\text{CC}}^{*} } \right)\) using theoretical thresholds, generating the category variable (CC), specified as follows:

$${\text{CC}}_{{{\text{kc}}}} = \left\{ {\begin{array}{ll} 1 \hfill & {{\text{if}}\,\,\,k_{0} \left\langle { {\text{CC}}_{{{\text{kc}}}}^{*} } \right\rangle k_{1} } \hfill \\ 2 \hfill & { {\text{if}}\,\,\,k_{1} \left\langle { {\text{CC}}_{{{\text{kc}}}}^{*} } \right\rangle k_{2} } \hfill \\ 3 \hfill & {{\text{if}}\,\,\,k_{2} \left\langle { {\text{CC}}_{{{\text{kc}}}}^{*} } \right\rangle k_{3} } \hfill \\ 4 \hfill & {{\text{if}}\,\,\,k_{3} \left\langle { {\text{CC}}_{{{\text{kc}}}}^{*} } \right\rangle k_{4} } \hfill \\ 5 \hfill & {{\text{if}}\,\,\,k_{4} \left\langle { {\text{CC}}_{{{\text{kc}}}}^{*} } \right\rangle k_{5} } \hfill \\ \end{array} } \right.$$

This variable (CC) can be stated in a more general form as:

$${\text{CC}}_{{{\text{kc}}}} = j \leftrightarrow k_{j - 1} < {\text{CC}}_{{{\text{kc}}}}^{*} \le k_{j} \quad j = 1, \ldots ,5$$

where the five k are the thresholds that divide each category in the ordered variable (Rodríguez-Puello et al. 2022). These parameters are strictly increasing in k, so that \(k_{0} = - \infty\) and \(k_{j} = + \infty\). We assume that the cumulative distribution function of the error term is normally distributed, allowing us to estimate an OPM in the form \(\Phi \left( \cdot \right)\) (Verbeke 2005). Therefore, the probability of observing the result j for the worker k can be defined as \(\Pr \left[ {{\text{CC}}_{{{\text{kc}}}} = j} \right] = \Phi \left( {{\text{CC}}_{{{\text{kc}}}}^{*} } \right).\) The parameters of this model were estimated via the maximum likelihood (ML) procedure.

Undoubtedly, the main empirical challenge is to reduce the bias associated with the potential effect of cultural supply on individual consumption of cultural goods, as city level of cultural supply (proxied by the share of cultural employment) is highly endogenous for two reasons. First, the existence of some local unobserved conditions that simultaneously affect cultural employment and cultural consumption, and second, and perhaps more importantly, the presence of reverse causality because a positive and significant association between city-level cultural supply and individual consumption of cultural goods might reflect simultaneously two facts: first, cities with a higher level of cultural supply foster individuals to consume more cultural activities and second, given the strong preferences for culture of individuals, cities provide more cultural supply.

To deal with this empirical issue, the literature suggests following an instrumental variable approach by using deeply lag measures as instruments to obtain a less biased estimated parameter (Garretsen and Marlet 2017). In Chile, however, since such instruments are not available, this paper employs an alternative method based on, partially, the shift-share analysis (Bartik 1991; Partridge and Rickman 2008; Moretti 2013). Specifically, we use the predicted 2017’s city share of cultural employment as an instrumental variable, which is computed according to the following expressionFootnote 8:

$$\widehat{{{\text{city}}\,{\text{share}}}}_{C,2017} = {\text{city}}\,\,{\text{share}}_{C,2000} \times r_{{\left( {2017 - 2000} \right)}} ,$$

where the predicted city share of cultural employment for city \(C\) in 2017 results from multiplying the 2000’s share of cultural employment for city \(C\) by the national cultural employment growth rate between 2000 and 2017 \((r_{{\left( {2017 - 2000} \right)}} )\). As stressed by Partridge and Rickman (2008), national employment growth should be exogenous to the city employment growth, therefore using the predicted city employment share as an instrumental variable ensures meeting the exclusion restriction whereas it also displays a high correlation with the endogenous variable. It is important to stress that, although our identification strategy provides a less-biased estimated coefficient, it is likely that some of the bias affecting our parameter of interest remains, therefore it should be interpreted cautiously.

5 Results

5.1 Aggregate cultural consumption

Table 2 presents the main results for aggregate cultural consumption using standard ordered probit models [columns (1) and (2)] and an instrumental variables approach [column (3)]. The estimates of Table 2 can be interpreted as the estimated marginal effects that an average individual consumes at least five cultural goods (CC = 5). In column (1), we include all the city-level control variables described in Table 1, while columns (2) and (3) include only city-level controls that can be treated as plausible exogenous. More specifically, precipitation and temperature are considered as exogenous as well as historical monuments. By contrast, population density, accessibility, and green areas are highly endogenous because they result from local conditions of cities. Although a common empirical approach to deal with endogenous variables is to use historical data as instrumental variables, for example, green areas from 50 years ago (Glaeser and Resseger 2010), data unavailability precludes us from using this approach, therefore, to test the reliability of the estimates, column (2) excludes these endogenous controls.

Table 2 Marginal effects on the probability of reporting the highest level of cultural consumption (consumption of cultural goods = 5)

As can be seen across all the columns of Table 2, there is a statistically significant positive association between the city-share of cultural employment and the probability of an individual to consume five or more cultural goods. These results suggest that workers––all else held constant––consume more culture if the city in which they live has a higher cultural supply, thereby cities specialized in cultural occupations experience a higher consumption of cultural goods. More specifically, according to the estimates of column (3), doubling the city share of cultural employment would increase the probability of reporting the highest level of cultural consumption by about 1%. As mentioned in Sect. 3, it is important to consider that this variable can capture variety of the cultural consumption as well as the intensity of its consumption.

The positive statistical association between cultural supply and consumption of cultural goods is rather robust to the exclusion of endogenous city-controls as shown in column (2). Likewise, the IV estimation method (column (3)) yields a quite similar estimated parameter, going from 0.008 to 0.009. While the first stage F-test is above the rule of thumb of 10, confirming that the instrument is strong enough, the null hypothesis of exogeneity is strongly rejected, with a p-value of 0.01, supporting the use of an IV approach. Therefore, IV results strongly suggest that cities with a greater supply of culture do significantly increase the individual consumption of cultural goods.

This positive association is also found when analyzing the effect of education on cultural consumption.Footnote 9 As specified in Table 2, across all the columns, the results show that years of education are positively associated with the probability of consuming cultural goods, which suggests that the higher educational attainment the stronger tastes for cultural goods, which might be associated to a differentiated taste for amenities depending on educational levels (Moretti 2013).

5.2 Disaggregate cultural consumption

While the previous analyses have studied the probability of consuming cultural goods by aggregating the nine categories of cultural goods, the following specifications address this analysis by focusing on the three categories of cultural goods separately using an IV approach. Table 3 shows the main estimates, column (1) presents the results for performing arts (theater, modern dance and ballet, opera, classical music concerts, circus performances), whereas the results for movie theaters are displayed in column (2), and finally column (3) presents the results for museums and art displays.

Table 3 Marginal effects on the probability of cultural consumption for every category: an instrumental variable approach

The main results are fully consistent with the previous estimates. The coefficient associated with the cultural supply has a positive and statistically significant effect on the consumption of two out of the three cultural goods. Interestingly, predicted city level of cultural employments turned out to be a strong instrument with a F-test well above the rule of thumb of 10. This result is valuable because it ensures that estimated coefficients do not lack precision due to a weak instrument. Despite this, it can be observed that the null hypothesis of exogeneity is not rejected in any models, with p-values ranging between 0.12 and 0.5, an empirical concern that will be addressed below. As in Table 2, years of schooling show a robust association with cultural consumption, specifically it shows a positive and statistically significant effect on the probability of an individual consumes any of the three categories of cultural goods.

Although previous results provide suggestive evidence in favor of a positive association between individual cultural consumption and city-level cultural employment for both aggregated and separate categories, one concern remains, which is the aggregated nature of the data set. As stressed by Ballas and Tranmer (2012), the study of social phenomena using microdata nested in different levels, such as individuals and cities, requires a particular approach that accounts for the multilevel character of the data sets. Consequently, we employ multilevel models to reduce any potential bias derived from the clustering of individuals into cities. This methodological approach has been used extensively in regional literature to study, for example, how individual and contextual factors affect population well-being (Ballas 2013; Ballas and Thanis 2022).

Table 4 summarizes the main results. The column (1) shows the estimates for aggregate cultural consumption, whereas the remaining three columns present the results for disaggregated categories. Eyeballing the estimated coefficients, we can see that they are qualitatively alike as those found in Tables 2 and 3. For example, a one percentage point increase in the cultural supply is associated with a 0.008 percentage point increase in the average marginal probability of aggregated cultural good consumption, whereas the IV estimate reported a coefficient of 0.009. For disaggregated cultural consumption categories, we do not find any statistical association for performing arts, whereas the positive association for the remaining two categories is confirmed, as found in Table 3. It important to notice, however, point estimates obtained from the multilevel approach are significantly lower than those in Table 3.

Table 4 Marginal effects on the probability of cultural consumption for every category: multilevel probit models

5.3 Assessing the causal link between city cultural supply and individuals’ cultural consumption

Previous analyses, based on an IV approach, suggested that an increase in cultural supply of cities rises an individual’s cultural consumption. However, to establish such a conclusion, an instrument should meet two conditions: relevance and exogeneity. While the relevance condition is easily tested using the first stage F-statistics, which showed to be well above the rule of thumb of 10, the results for disaggregate cultural consumption casts some doubt on the exogeneity of the employed instrument. Although the exogeneity condition cannot be empirically tested, what we can do is to test the robustness of the IV estimates by relaxing the exclusion restriction of the IV.

Based on the method proposed by Conley et al. (2012) and implemented by Clarke and Matta (2018),Footnote 10 we allow for a direct effect of the instrument on the individual cultural consumption (i.e., a violation of the exclusion restriction) within a range of values. These priors are assumed to be the zero and OLS estimators resulting from regressing cultural consumption on the instrumental variable for both aggregate and disaggregate individual cultural consumption. Consequently, the direct effect of the instrument on the dependent variable lies between \(\left[ {0, \,\,\hat{\gamma }_{{{\text{OLS}}}} } \right]\), while we first did not assume any distributions for \(\hat{\gamma }\)—an approach called the union of confidence intervals (UCI), after we assumed that \(\hat{\gamma }\) follows a normal distribution, that is local to zero (LTZ) approach.

Figure 1 summarizes the main results. Each figure shows the 95% confidence interval for the IV estimates under several degrees of endogeneity: from zero to several values for \(\hat{\gamma }_{{{\text{OLS}}}}\). As can be seen, for aggregate cultural consumption (Panel A), the positive effect of city-cultural supply tolerates a strong degree of endogeneity, it loses its statistical significance only for larger values of \(\hat{\gamma }_{{{\text{OLS}}}}\). Likewise, for movie theater (Panel C), the causal effect of city-cultural supply is robust to a large degree of endogeneity, whereas a slight less robust causal effect is observed for museum and art displays. Finally, Panel D confirms the results found in Table 3, there is no causal effect of city-cultural supply on individual consumption of performance arts irrespective of the proposed degree of endogeneity.

Fig. 1
figure 1

Assessing the causal effect of city cultural supply on cultural consumption. Notes Each graph shows the 95% confidence interval (CI) estimates for the impact of the instrument on the four different dependent variables using the UCI and LTZ methods. Dependent variables for each column: (1) consumption of cultural goods; (2) performing arts: theater, modern dance and ballet, opera, classic music concerts, and circus; (3) movie theater; and (4) museum and arts displays. Only exogenous control variables are included in the models. The solid line is the CI with the UCI method. The dashed line is the CI with the LTZ method, by assuming \(\gamma\) is normally distributed

Overall, this analysis tends to confirm that, at least part of the estimated positive association, is due to the existence of causal effect for aggregate consumption as well as for two disaggregated categories: movie theater and museum and art displays. This result is found even for a large degree of endogeneity. However, a quite different scenario characterizes the consumption of performing arts because the evidence does not allow us to support the existence of causal effect between its consumption and city-cultural supply. Apparently, consumption of performing arts is mainly driven by individual characteristics such as schooling and cultural capital rather than the cultural environment that cities can offer to their residents. For the remaining two cultural goods, however, city-cultural supply seems to play a major role to boost their consumption.

6 Conclusions

In this paper, we have taken a step back compared to most studies in the urban economy literature and directly test whether—after controlling for individual characteristics—city-level cultural supply is associated with spatial differences in the probability of consuming cultural goods. To meet this objective, we employed individual data of cultural consumption for 2017 along with several attributes of the Chilean cities.

The main results confirm that cities’ cultural employment shape workers’ decisions to consume cultural goods in Chile. The positive effect of cultural supply of cities on individual consumption of cultural goods is found for aggregate cultural consumption as well as separated categories of cultural goods. IV estimates suggest that doubling the city level of cultural supply would increase the estimated marginal effects that an average individual consumes at least five cultural goods by 1%. A larger positive effect is found when focusing on each good separately. In addition to this, the main findings seem to confirm that schooling is an important individual trait associated with the consumption of cultural goods in Chile—a result that was comparatively robust across all the specifications.

When testing the reliability of the main results, we allow for the instrument to be correlated with the residuals using the method proposed by Conley et al. (2012). The results confirm that IV estimates tolerate a high degree of the violation of the exclusion restriction, therefore—except for performing arts—we find convincing evidence of the positive effect of cultural supply of cities on individual cultural consumption.

One of the main contributions of this study is to provide empirical evidence on a lower-income country, where studies about cultural participation and its determinants are scarce compared with Europe and the USA, especially with a regional dimension. Since city-cultural supply enhances significantly cultural consumption in lower-income countries, it is expected this effect to be larger for developed economies because tastes for amenities is positively related to the level in income of the country (Partridge 2010). Also, it is important to highlight that these findings are valuable, not only for scholars, but also for policymakers—particularly in Chile, a country with persistent levels of spatial inequality. In this regard, the main results of this study suggest that a biased public allocation of cultural infrastructure towards lagging cities coupled with additional local policies oriented to improve the quality of life of residents—could create the right incentives for fostering, for example, a more equal distribution of skilled workers across the country.

Although this study provides valuable empirical evidence about the importance of cities’ cultural supply, it is not without limitations. The main limitation is related to the data availability, as Chile has only one data set on cultural consumption, which precludes researchers from performing a panel data analysis or estimating a demand system model given the lack of information on prices. Additionally, the available data used in this study are representative of the 85% of the total target population, and since part of the sample is lost during data preparation, we interpret the results with caution, treating them as representative only for the selected sample. In the following years, futures studies should address these empirical tasks along with some additional analyses about the effect of city-cultural supply on the attraction of human capital as Backman and Nilsson (2018) assessed for Sweden.