Islam and democracy

Khodaverdian, Saeed Article — Published Version Islam and democracy Kyklos Provided in Cooperation with: John Wiley & Sons Suggested Citation: Khodaverdian, Saeed (2022) : Islam and democracy, Kyklos, ISSN 1467-6435, Wiley, Hoboken, NJ, Vol. 75, Iss. 4, pp. 580-606, https://doi.org/10.1111/kykl.12315 This Version is available at: https://hdl.handle.net/10419/266760 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by-nc-nd/4.0/ ORIGINAL ARTICLE Islam and democracy Saeed Khodaverdian Correspondence Saeed Khodaverdian, University of Hamburg, Economics Department, Von-Melle-Park 5, 20146 Hamburg, Germany. Email: [email protected] Abstract The Muslim countries have a lower democracy level on average than the other countries. Yet, this cross-country comparison is not sufficient to blame Islam for the democracy deficit. This issue is related to the numerous fixed characteristics that differ across the countries and that also have an effect on democracy. When the differences are not considered, the estimates mix the effect of the fixed characteristics with the effect that Islam has on democracy. We show this issue for a large set of countries covering the period 1950–2015. In cross-country comparisons, we find a negative link between the share of Muslims in the total population and common measures for democracy. Yet, when the fixed characteristics are considered, the negative link vanishes. Thus, the lower democracy level has to be due to a fixed characteristic; it cannot be associated to (timevarying) Muslim shares. Numerous empirical specifications support this finding. We also measure Islam by the average per-day length of fasting during the month of Ramadan exploring whether the intensity of practicing the core teachings of Islam has a negative effect on democracy. Again, the evidence rejects a negative effect of Islam on democracy. If anything, the effect is positive. The findings are important for social coexistence. Received: 31 October 2021 Revised: 19 April 2022 Accepted: 6 June 2022 DOI: 10.1111/kykl.12315 This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. © 2022 The Author. Kyklos published by John Wiley & Sons Ltd. 580 Kyklos. 2022;75:580–606. wileyonlinelibrary.com/journal/kykl 1|INTRODUCTION Today, around half of the countries in the world are non-democratic. There are different explanations for this pattern, yet, all of them are disputed. Only when it comes to Islam, there is widespread agreement that the religion has a negative effect on democracy. In this study, we challenge this view. Our empirical analysis shows that the negative association between Islam and democracy exists only in cross-country comparisons. Once country fixed effects are included, the negative link vanishes. The relationship between Islam and democracy is discussed extensively in the literature (e.g. Lipset (1994), Midlarsky (1998), Barro (1999), Ross (2001), Karatnycky (2002), Fish (2002), Donno and Russett (2004), Pryor (2007), Borooah and Paldam (2007), Rowley and Smith (2009), Facchini (2010), and Potrafke (2012,2013)). 1 The conclusion of these efforts is that Muslim countries are significantly less associated with democracy than other societies. According to Maseland and Van Hoorn (2011, p. 483) this notion “... is readily accepted.”Yet, the crosscountry link is not sufficient to blame Islam for the democracy deficit. The reason is related to the numerous fixed characteristics that differ across the countries. For example, Acemoglu et al. (2001,2002) show that colonial history has had significant influence on the type of institutions implemented by the Europeans and that this feature is key for the political orientation of the countries today. Similarly, the geographical position of the country is an important determinant for democracy (e.g. Persson and Tabellini (2009) and Acemoglu et al. (2019)). These features are not caused by Islam. Thus, the empirical model has to consider them otherwise the estimates mix the effect of Islam with the effects that these fixed characteristics have on democracy. Figure 1illustrates our point. It displays the development of two of the most common measures for democracy. The measures are normalized between 0 and 1 whereby higher values indicate a higher level of democracy (details are discussed in the data section). It is clearly visible that across the countries, the democracy index decreases as the share of Muslims in the total population increases. The negative cross-country link exists in almost any year between 1950 and 2015 (sample period). However, the conclusion changes when the development of democracy is considered over time. From this perspective, it is clearly visible that democracy has different initial levels, yet, it develops very similar for all the countries irrespective of the Muslim share. The empirical analysis confirms the different views. In cross-country comparisons, the Muslim share is negatively associated with the measures for democracy. However, in within-country regressions, the negative link vanishes. We provide ample evidence showing the robustness of the pattern. For example, we find that the relationship is not driven by endogeneity. This is an important point because it clarifies that our estimates actually show the causal effect that runs from Islam to democracy. That effect is also not driven by the fact that most countries in the world have a minor Muslim share where Islam plays no important role for democracy. Our estimates are also robust to the criticism that the inclusion of country fixed effects removes a large fraction of the variation that is available across the countries. To this end, it is worth highlighting that some of our estimates are significant. Yet, they are positive. In addition, we perform a long-term specification where we consider the change in the measures for Islam and democracy over the sample period. As shown in Figure 2, there is meaningful variation in the long-term. Yet, there is no evidence for a negative effect. 2 1 Numerous variables have been suggested for the negative cross-country link including the Arab culture, Israeli-Palestine conflict, lack of gender equality, natural resources, and restriction of religious freedom. Fixed country characteristics are not considered. 2 Figure 2displays three plots. The plot at the top shows a positive relationship between the change in the Muslim share and the change in the Political Rights index. Yet, it also highlights two outliers, Qatar and the United Arab Emirates (UAE). In Qatar, the average Muslim share between 1972 and 1982 (beginning of the sample) was 96% and it went to an average of 47% between 2005 and 2015 (end of the sample). This corresponds to a change of 49 percentage points. In the UAE, the Muslim share went from an average of 90% to an average of 67% (=23 percentage points) over the same periods. In the middle plot, the two outliers are excluded. The plot at the bottom is based on the Polity 2 index with averages over 1950–1960 (beginning of the sample) and averages over 2005–2015 (ending of the sample). Despite another period, it confirms the non-negative relationship. See Section 4.4 for more details. KHODAVERDIAN 581 In the analysis, we follow the literature by measuring Islam by the Muslim share. Although plausible, this measure might miss the point. When religions are concerned, the message of God is passed to the people. In this setting, it is plausible assuming that it is not the people but the messages (i.e. the teachings of the religion) that have an effect on outcomes. If so, measuring Islam by the Muslim share is meaningless. We address this issue by measuring Islam by the average per-day length of fasting during the month of Ramadan. Below, we explain the practice and outline our motivation for it. We also discuss the related literature working out that the measure includes the core teachings of Islam and that its variation affects Muslims' believes and actions. Since the length of Ramadan is an exogenous event, this measure too shows the causal effect that runs from (the teachings of) Islam to democracy. The application of the measure shows no evidence for a negative effect within the Muslim-majority countries. In sum, the evidence rejects the idea that Islam has a negative effect on democracy. If anything, the religion supports democracy within the Muslim countries. We observe this pattern by some of the estimates turning out significantly positive. Although this might appear to be due to noise in the data or due to misspecifications in the FIGURE 1 The figures show the development of democracy for countries grouped at different shares of Muslims in the total population. In the upper plot, democracy is measured by the Political Rights index normalized between 0 and 1 (higher values indicating more democracy) and in the lower plot, democracy is measured by the Polity 2 index normalized between 0 and 1. The data sources are named in Section 2. 582 KHODAVERDIAN FIGURE 2 Legend on next page. KHODAVERDIAN 583 empirical model, it is worth highlighting that a positive effect is in line with the (non-macro) literature. 3 For example, Islamic decision-making is based on the concepts of consultation (shura), legal consensus (ijma), and independent reasoning (ijtihad). These principles advise Muslims to counsel each other in affairs concerning political, economic, as well as social life seeking the agreement of the community (ummah). The elements ensure that neither power becomes concentrated in the hands of a few (e.g. Voll and Esposito (1994)) nor will majority-voting be rejected (e.g. Farooq & Aktaruzzaman, 2018). This setting resembles that of an ideal democracy (e.g. Esposito et al., 1996). There is also a significant micro-literature showing that Muslims have similar or even higher demands for democracy than non-Muslims (e.g. Rose (2002), Tessler (2002a,2002b), Hofmann (2004), Tessler and Gao (2005), Jamal and Tessler (2008), Rowley and Smith (2009), Shafiq (2010), Farooq and Aktaruzzaman (2018), and Tessler et al. (2012)). Our results confirm this pattern by showing estimates that are either insignificant (similar demand) or significantly positive (higher demand). This is particularly true in countries where Muslims represent a majority. Yet, most of our estimates are insignificant. This pattern is often criticized for being less informative. As discussed by Abadie (2020), this is not true in general. In cases with a strong prior for an outcome, insignificant estimates carry more information because they induce a larger change in beliefs. This applies particularly to the relationship between Islam and democracy where a negative association is prevalent. Moreover, we explain Islam's democracy paradox (asking why the Muslim countries have so little democracy even though the Muslims like democracy so much, e.g. Rowley and Smith (2009)). As we show, this is related to the fixed characteristics of the countries that are either unchangeable (e.g. geography) or change only over the very long period (e.g. national identity). In the conclusion, we return to this discussion. The analysis also contributes to the literature that challenges common wisdom based on insignificant estimates. In general, Acemoglu et al. (2008) reject the idea that higher income leads to more democracy. With regards to Islam, Jetter and Stadelmann (2019) find no support that Muslims are associated with more terror per-capita. 4 We discuss the implications of our findings in the conclusion. 3 There are few macro-studies that show a positive effect of Islam on components of democracy. For example, Clingingsmith et al. (2009) show that the performing of the Hajj pilgrimage to Mecca (one of the five pillars of Islam) increases both belief in religious freedom and sympathy for female education and female employment. Similar results are reported by Meyersson (2014) showing that Islamic rule increased female education in Turkey. 4 There are also studies conducted for individual cities/countries. For example, Cesarini et al. (2016) show that wealth is irrelevant for health outcomes in Sweden. Meghir et al. (2018) show that education does not improve the mortality rate in Sweden. The prior that going to an elite school leads to higher test scores is also not supported by the data (see Abdulkadiroglu et al. (2014) for schools in Boston and New York, and Lucas and Mbiti (2014) for schools in Kenya). De Ree et al. (2018) show that increasing teacher salary has no impact on test scores in Indonesia. There is also no support that migrant agricultural laborers drive down wages for U.S. workers (see e.g. Card (1990) and Clemens et al. (2018)). FIGURE 2 The plots display the changes in averages for the measures for Islam and democracy over the sample period. Islam is measured by the %-share of Muslims in the total population. Democracy is measured by two figures including the Political Rights index and Polity 2 index each normalized between 0 and 1 (higher values indicating more democracy). The averages are calculated over the first and last 11 years in each sample. Due to data limitations, the averages in the top and middle plots are taken over the period 1972–1982 (for the beginning) and 2005–2015 (for the ending). In the plot at the bottom, the averages are taken over the periods 1950–1960 and 2005–2015. In the top plot, Qatar and the United Arab Emirates (UAE) identify as outliers. The middle plot repeats the analysis of the top plot excluding the two countries. See also Footnote 2 and Section 2 for more information. 584 KHODAVERDIAN 2|DATA AND DESCRIPTIVE STATISTICS The analysis utilizes data from several sources. For democracy, we employ the Political Rights index from Freedom House and the Polity 2 index from the Polity IV Project (Marshall et al., 2017). The measures are widely used in the literature since they evaluate democracy from different perspectives. The Political Rights index measures democracy from the perspective of the citizens. It is based on a catalogue of 25 questions concerning the electoral process, political pluralism and participation, functioning of government, freedom of expression and beliefs, associational and organizational rights, rule of law, as well as personal autonomy and individual rights. The Polity 2 index is a revised 21-scale measure that gauges democracy based on competitiveness of executive recruitment, openness of executive recruitment, constraints on the chief executive, regulation of participation, and competitiveness of participation. Both measures are available on an annual basis. However, their coverage across countries and years varies. For example, the Political Rights index is available for a larger set of countries, yet, data is given only from the year 1972 onwards. The Polity 2 index includes information for fewer countries, yet, data begins in the year 1800. Another difference between the two indices is that they vary in range. To have comparable estimates, we follow the literature (e.g. Acemoglu et al. (2008) and Kotschy and Sunde (2017)) by normalizing each index to range between 0 (full autocracy) and 1 (full democracy). The second data source concerns the measure for Islam. Our main proxy is the share of Muslims in the total population. We obtain this information from the World Religion Project maintained by Maoz and Henderson (2013). The database offers historical records for almost all countries in the world between 1945 and 2010. The advantage of the dataset is its large coverage across both countries and years. However, its disadvantage is that it is not balanced (i.e. the Muslim share is not given for all countries at any time). Moreover, the estimates are provided on a five-year basis (i.e. for 1945, 1950,…, 2010). We interpolate the data to get a panel with annual observation covering the sample period 1950–2015. In some specifications, we focus on countries where Muslims represent a vast majority in the total population. We evaluate this feature by a threshold of ≥75% Muslims in the total population. Countries that meet this criterion are referred to as Muslim-majority countries. Our alternative measure for Islam is the average per-day length of fasting during the month of Ramadan. We make a few simplification assumptions for the construction of this measure: (i) we assume Ramadan lasts exactly 30 days in each year (in reality it can last 29 or 30 days depending on the seeing of the new moon), (ii) we assume that on the days of Ramadan, fasting begins with the astronomical sunrise and ends with the astronomical sunset (in reality the Quran specifies that fasting begins at dawn, i.e. first light), (iii) we assume that the capital is representative for the entire country. Based on these assumptions, we proxy Ramadan Fasting as the average per-day length of daylight in the capital of a country over a 30-day period beginning with the first day of Ramadan. We construct this measure based on information from the Islamic Philosophy Online webpage (www. muslimphilosophy.com) where we convert the dates from the Islamic calendar to the Gregorian calendar. For each day of Ramadan, we observe the corresponding sunrise and sunset times using information provided by the Astronomical Applications Department of the U.S. Naval Observatory (www.aa.usno.navy.mil). The webpage requires geographic coordinates to determine sunrise and sunset times. Data for the geographic coordinates of the capitals (i.e. longitude and latitude) is provided by the French institute for research into international economics (www.cepii.fr). These sources enable the construction of our Ramadan measure for a given country and year. We re-scale the Ramadan measure to decimal values, i.e. a value of 10.5 in our dataset means 10 hours and 30 minutes. Finally, we consider two control variables concerning economic growth and development. We measure the former by GDP per-capita. In detail, we obtain data for total GDP at chained PPP (in 2011 US-$) dividing it by the total population size, both available in the Penn World Tables (Feenstra et al., 2015). This database covers the entire sample period. For development, we consider life expectancy at birth (in years). Data is KHODAVERDIAN 585 gathered from the World Indicators Database (WDI) maintained by the World Bank. Albeit, the data begins in 1960. In sum, our baseline sample contains annual information for a maximum of 189 countries covering the period 1950–2015. However, the sample is not balanced meaning that we do not have data for all the countries in all the years. Thus, depending on the model specifications, the sample size varies. Table 1displays the summary statistics for our sample. At the top of the table, Panel A shows information for the entire baseline sample. In the two panels beneath, we divide the baseline sample according to the 75%-threshold showing separated data for the Muslim-majority countries. This provides further insights. Overall, the world has an average democracy index of 0.54. This is shown by both measures for democracy. However, there is a significant difference with respect to the Muslim share. In the Muslim-majority countries, the democracy index is only about half that large than in the other countries. This pattern is confirmed by both measures for democracy. With respect to economic growth though, the countries in the two groups are on average almost equally rich (shown by average GDP per-capita of 10,815 vs 11,066). In terms of human development in turn, the Muslim-majority countries lag behind since life expectancy at birth is on average slightly lower for them (60.38 vs 64.28). When Ramadan fasting is concerned, the average per-day length of fasting in the Muslim-majority countries is 12.13 normalized hours. TABLE 1 Islam and democracy: summary statistics Variable Observations Mean Std. Dev. Min Max Panel A: Baseline sample Polity 2 index (normalized) 8,604 0.54 0.37 0 1 Political Rights index (normalized) 7,587 0.54 0.37 0 1 Muslim share (%-share of total population) 9,718 0.24 0.36 0 1 GDP per-capita (at chained PPP, in 2011 US-$) 8,440 10,859 16,027 223 251,882 Life expectancy (at birth, in years) 8,733 63.55 11.48 18.91 85.42 Panel B: Countries with less than 75% Muslims in the total population Polity 2 index (normalized) 6,936 0.60 0.36 0 1 Political Rights index (normalized) 6,212 0.61 0.37 0 1 GDP per-capita (at chained PPP, in 2011 US-$) 6,954 10,815 13,477 223 153,458 Life expectancy (at birth, in years) 7,084 64.28 11.56 18.91 85.42 Panel C: Countries with at least 75% Muslims in the total population Polity 2 index (normalized) 1,668 0.29 0.29 0 0.95 Political Rights index (normalized) 1,375 0.27 0.24 0 1 GDP per-capita (at chained PPP, in 2011 US-$) 1,486 11,066 24,684 482 251,882 Life expectancy (at birth, in years) 1,649 60.38 10.59 28.12 77.05 Ramadan Fasting length (normalized hours per day) 1,748 12.13 1.25 9.12 15.09 Notes: The table shows summary statistics for the main variables. The Polity 2 index is from the Polity IV project (Marshall et al. (2017)) and the Political Rights index is from Freedom House. Both indices are normalized between 0 (absolute autocracy) and 1 (absolute democracy). Data on the share of Muslims in the total population is from the World Religion Project (WRP) maintained by Maoz and Henderson (2013). Data on GDP per-capita (determined by GDP divided by total population size) is from the Penn World Tables (Feenstra et al. (2015)). Data on life expectancy is from the World Development Index (WDI) database maintained by the World Bank. Ramadan fasting refers to the average time from sunrise to sunset times for a 30-day period beginning with the month of Ramadan in a given year and for a given capital. Panel A shows statistics for the baseline sample. Panels B and C show separate statistics for countries with different Muslim shares in the total population. More information is given in Section 2. 586 KHODAVERDIAN 3|EMPIRICAL MODEL We estimate the effect of Islam on democracy based on the following model: dct ¼αdct1þγIct1þβXct1þμcþτtþuct ð1Þ Here, drepresents the level of democracy in country cin yeart. The variables on the right side explain that level. They include the lag of democracy (dct1), a lagged measure for Islam (Ict1), a lagged vector (Xct1) for control variables, full sets of country and year fixed effects (μcand τt), and the error term (uct). Throughout the analysis, we report standard errors robust against arbitrary heteroskedasticity and serial correlation at the county-level. Democracy is a persistent process also showing mean-reverting dynamics. The literature recommends to consider these issues by including the lag of democracy on the right side of the equation (e.g. Acemoglu et al. (2008) and Kotschy and Sunde (2017)). Although justified, the dynamic model takes out a large fraction of the variation in democracy. This leaves little variation to be explained by the other variables of the model. We tackle this issue by employing a static version of the empirical model (where α¼0) as well as a dynamic version (where α≠0). Another estimation issue concerns the time structure of the panel. In detail, it is unlikely that changes in the measures for Islam have immediate effects on democracy. Since the timing is unclear, we consider three different data frequencies: one-year (where data is observed every year), five-year, and ten-year. Depending on the frequency, the first lag (t1) refers either to the value one year ago, five years ago, or ten years ago. Our main interest lies on the effect of Islam on democracy. We model that variable by Ict1. Thus, γis the main estimate of interest. When γis significant and positive (negative), it follows that Islam has a positive (negative) effect on democracy. However, when γis insignificant, the estimate suggests that Islam is irrelevant for democracy. As discussed above, the type of analysis (cross-country or within-country) is important to estimate the true effect of Islam on democracy. The empirical model considers this issue by the country fixed effects, μc. In crosscountry regressions, the (country and year) fixed effects are excluded from the model. In within-country regressions though, they are all included in the model. Comparison of the results provides deep insight for the features that keep the Muslim countries less democratic. 4|MUSLIM SHARE AND DEMOCRACY We begin the analysis by measuring Islam by the share of Muslims in the total population. The choice for this measure follows from the claim that the larger the Muslim share the more are Islamic views expressed in political outcomes. 4.1 |Main Findings Tables 2and 3represent our main findings. They show that the negative link between the Muslim share and common measures for democracy exist only in cross-country comparisons. Once country fixed effects are included in the model, the negative link vanishes. In Table 2, democracy is measured by the Political Rights index. As mentioned in Section 3, we employ both a static version as well as a dynamic version of the empirical model (named in the Panel-head). Moreover, we present estimates for both types of analysis including cross-country comparisons and within-country regressions. This is specified in the top row following Panel A. The second-top row names the frequency at which the data is observed. The specifications in columns (1) to (3) show that there is a stark negative cross-country link between the Muslim share and democracy. The columns also clarify that the pattern is independent of the data frequency. To specify one KHODAVERDIAN 587 TABLE 5 Islam and democracy: omitted variable bias Dependent variable: democracy Measure for democracy Political Rights index (normalized) Polity 2 index (normalized) Empirical model Static Dynamic Static Dynamic (1) (2) (3) (4) (5) (6) (7) (8) Democracy t-1 0.422*** (0.0382) 0.421*** (0.0383) 0.557*** (0.0375) 0.536*** (0.0385) Muslim share t-1 0.301 (0.216) 0.294 (0.214) 0.181 (0.135) 0.181 (0.135) 0.378* (0.208) 0.442** (0.176) 0.306*** (0.105) 0.299*** (0.104) (log) GDP per-capita t-1 0.00741 (0.0237) 0.00709 (0.0239) 0.000793 (0.0167) 0.000562 (0.0168) 0.0298 (0.0238) 0.0276 (0.0230) 0.0205 (0.0139) 0.0201 (0.0149) (log) Life expectancy t-1 0.104 (0.146) 0.0309 (0.0965) 0.265 (0.178) 0.0586 (0.0970) Country fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Observations 1,391 1,375 1,224 1,211 1,557 1,450 1,495 1,396 R-squared 0.126 0.126 0.283 0.283 0.305 0.325 0.544 0.539 Countries in sample 166 164 166 164 151 151 151 151 Notes: The table shows OLS estimates for the conditional within-country correlation between the Muslim share and two measures of democracy. The top row informs about the measure for democracy and the second-top row specifies the empirical model. All regressions are based on a five-year data frequency. Consequently, the subscript t1 refers to the value five years ago. Columns with odd numbers include the first lag of (log) GDP per-capita as control variable. Columns with even numbers add the first lag of (log) life expectancy on top. Standard errors (reported in parentheses) are robust against heteroskedasticity and serial correlation at country-level. Significance levels are. * * * p < 0.01, ** p < 0.05, * p < 0.1. 594 KHODAVERDIAN dct ¼αdct1þIct1γgþγmsc  þμcþτtþuct For countries where sc¼0, the general term γgestimates the effect of the Muslim share on democracy and for countries where sc¼1, the sum of γgand γmshows the effect of Islam on democracy. In the analysis, we set sc¼1 for countries that have ≥75% Muslims in the total population, and 0 otherwise. This threshold balances the trade-off between capturing the true effect of Islam and still allowing a fair distribution of the data. Specifications with a 50%- threshold did not lead to other results than those shown here (available upon request). In Table 6, we apply both methods to explore country heterogeneous effects. The top row of the table clarifies the measure for democracy, the second-top row names the approach (separated model and interacted model), and the third-top row specifies the empirical model (static and dynamic). When the interacted model is applied, the table explicitly displays the effect in the Muslim-majority countries (called MM-countries in the table to save space). TABLE 6 Islam and democracy: country heterogeneity Measure for democracy Dependent variable: democracy Political Rights index (normalized) Polity 2 index (normalized) Approach Separated model Interacted model Separated model Interacted model Empirical model Static Dynamic Static Dynamic Static Dynamic Static Dynamic (1) (2) (3) (4) (5) (6) (7) (8) Democracy t-1 0.324*** (0.0985) 0.427*** (0.0371) 0.477*** (0.124) 0.586*** (0.0347) Muslim share t 0.407 (0.529) 0.0736 (0.400) 0.188 (0.396) 0.0759 (0.249) 1.078** (0.496) 0.764* (0.400) 0.117 (0.255) 0.150 (0.148) Muslim share t-1 *s c 0.0726 (0.121) 0.0621 (0.0805) 0.171** (0.0776) 0.0826 (0.0525) Effect in the MM-countries 0.261 (0.278) 0.138 (0.175) 0.288 (0.194) 0.233* (0.118) Country fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Observations 281 248 1,478 1,344 341 316 1,667 1,647 R-squared 0.050 0.139 0.121 0.280 0.289 0.443 0.302 0.553 Countries in sample 36 36 189 189 35 35 161 161 Notes: The table shows OLS estimates for the within-country correlation between the Muslim share and two measures for democracy in countries where Muslims represent ≥75% of the total population (referred to as Muslim-majority [or MM-]countries). The top row informs about the measure for democracy and the second-top row specifies the approach how the MM-countries are identified. In columns (1)–(2) and (5)–(6), the MM-countries are separated from the other countries running isolated regressions only for them. In columns (3)–(4) and (7)–(8), the MM-countries are identified by a 75%-dummy interacted with the Muslim share. In this case, the addition of the two individual estimates for the Muslim share shows the overall effect in the MM-countries (see Section 4.3 for details). The third-top row informs about the empirical model. All regressions are based on a five-year data frequency. Consequently, the subscript t1 refers to the value five years ago. Standard errors (reported in parentheses) are robust against heteroskedasticity and serial correlation the country-level. Significance levels are. * * * p < 0.01, ** p < 0.05, * p < 0.1. KHODAVERDIAN 595 Some of the specifications show evidence for country heterogeneity. Yet, the evidence speaks in favor of a positive effect. In detail, when democracy is measured by the Political Rights index, all estimates are insignificant. We observed the same pattern in the main findings. However, when the Polity 2 index is employed, it turns out that the positive effect of the Muslim share on democracy shown in the main findings is basically driven by the Muslimmajority countries. In column (5), the coefficient for the Muslim share is 1.078 (0.496). This is more than three times the average effect that applies to all countries (0.313 shown in Table 3, Panel A, column (5)). The coefficient remains large even when lagged democracy is considered. This is shown in column (6) where the coefficient is 0.764 (0.400). This is also three times the corresponding coefficient for the average effect across all countries (0.258 shown in Table 3, Panel B, column (5)). The pattern becomes slightly less clear when the interaction model is applied. The marginal effect of the Muslim share is significant, 0.171 (0.0776), yet, the general effect is too imprecisely estimated, 0.117 (0.255). This makes the estimate for the effect of the Muslim share on democracy in the Muslim-majority countries insignificant, 0.288 (0.194). However, when the dynamics of democracy are considered, the estimate in column (8) shows that the positive average effect shown in Table 3is driven by the Muslim-majority countries. The exercises in Table 6are insightful because they show that the main findings apply to all countries irrespective of whether Muslims represent a minority or majority in the total population. If anything, a larger Muslim share supports democracy in the Muslim-majority countries. 4.4 |Long-Term Effect The main findings assume that Islam shows its effect on democracy in a short period. The specifications in Tables 2 and 3consider a time gap of up to 10 years between the Muslim share and the measures for democracy. Yet, it is possible that Islam shows its true effect over longer periods. Here, we explore this issue showing that even over a half-century period there is no evidence for a negative effect of Islam on democracy. When exploring the long-term effect of the Muslim share on democracy, we deviate from the five-year data frequency. Instead, we consider the changes in the averages of the measures for Islam and democracy over the sample period. In detail, for each measure, we determine the average over the initial 11-year period for which data is given. Then, we do the same with the last 11-year period for which data is given. Finally, we calculate the change as the difference between the last period minus the initial period. This approach leads to a sample with only one observation per country. Yet, it enables the estimation of the long-term effect of Islam on democracy over the sample period. The plots in Figure 2illustrate this specification. They suggest that there is no negative effect. Table 7supports that conclusion by showing the estimates for the long-term specifications. The top row of the table informs about the measure for democracy, the second-top row names the countries that are included in the sample, and the thirdtop row specifies the empirical model (static and dynamic). Columns (1) and (2) show a specification where all countries are included in the sample. Irrespective of whether initial democracy is considered, the estimates turn out insignificant suggesting that there is no effect of the average Muslim share on average democracy over the period 1972– 2015. This effect might be biased by the two outliers (Qatar and the UAE) shown in the top-plot of Figure 2. Without these countries, the slope of the regression line becomes flatter in the middle-plot. We observe the same pattern in columns (3) and (4) where the two outliers are excluded from the sample. In the static model, the coefficient becomes less than half compared to the case in column (1) and in the dynamic model when initial democracy is considered, the estimate even changes sign becoming negative. Albeit, in both cases, the estimates are insignificant by wide margins. The same insignificant pattern emerges in columns (5) and (6) where democracy is measured by the Polity 2 index. 8 8 In Table 7, initial democracy turns out negative. Note that this is not a misspecification. The pattern suggests that the change in democracy over the sample period is smaller the higher initial democracy was at the beginning of the sample period. In unreported specifications, we found that dependingon the measure for democracy, the change in sign occurs after 20 to 30 years. Acemoglu et al. (2008) observe the same pattern in their Tables 2and 3after a period of 20 years. 596 KHODAVERDIAN In sum, the specifications in Table 7show that there is also no evidence for a negative effect of Islam on democracy over a half-century period. 5|RAMADAN FASTING AND DEMOCRACY So far, Islam is measured by the Muslim share. As mentioned above, this measure follows the claim that the larger the share of Muslims in the total population the more are Islamic views expressed in political outcomes. Although this is plausible, it is similarly possible that the true effect of Islam on democracy does not run through the scope of its followers but rather through its core teachings. In this case, the Muslim share is irrelevant for decision-making and it does not represent an appropriate measure for Islam. In this section, we explore this issue. In detail, instead of the Muslim share, we measure Islam by Ramadan Fasting. Since this feature is not known in general, we briefly explain that practice also highlighting its advantages for the empirical analysis. TABLE 7 Islam and democracy: long-term effects Dependent variable: change in democracy Measure for democracy Political Rights index (normalized) Polity 2 index (normalized) Sample All countries All but Qatar & UAE All countries Empirical model Static Dynamic Static Dynamic Static Dynamic (1) (2) (3) (4) (5) (6) Initial democracy (;19721982) 0.352*** (0.0558) 0.365*** (0.0576) Change in average Muslim share (;19721982 to ;20052015) 0.398 (0.285) 0.304 (0.306) 0.175 (0.534) 0.172 (0.487) Initial democracy (;19501960) 0.650*** (0.0667) Change in average Muslim share (;19501960 to ;20052015) 0.269 (0.369) 0.0803 (0.255) Country fixed effects No No No No No No Year fixed effects No No No No No No Observations 162 162 160 160 103 103 R-squared 0.007 0.178 0.001 0.182 0.004 0.451 Countries in sample 162 162 160 160 103 103 Notes: The table shows OLS estimates for the correlation between country-specific changes in the averages of the Muslim share and two measures for democracy over the sample period. In columns (1)–(4), the averages are taken over the periods 1972–1982 (sample beginning) and 2005–2015 (sample end). In columns (5)–(6), the averages are taken over the periods 1950–1960 (sample beginning) and 2005–2015 (sample end). The top row informs about the measure for democracy and the second-top row specifies the countries in the sample. The third-top row names the empirical model. Standard errors (reported in parentheses) are robust against heteroskedasticity and serial correlation the country-level. Significance levels are. * * * p < 0.01, ** p < 0.05, * p < 0.1. KHODAVERDIAN 597 5.1 |Ramadan Fasting as Measure for Islam Ramadan is the name of the ninth month of the Islamic calendar. According to Islamic theology, the month is holy because the first verses of the Quran were revealed to Prophet Muhammad during this period. In commemoration of the event, Muslims observe Ramadan as a month of fasting, prayer, reflection, alms giving, and community. 9 Fasting in Islam means that all types of consumption are stopped from sunrise (fajr) until sunset (maghrib) for the days of Ramadan. Instead, religious values are put at the center of daily routines. Thus, fasting includes the core teachings of Islam. The importance of the fast can also be inferred from the fact that Ramadan Fasting represents one of the five pillars of Islam (next to faith (shahada), prayer (salah), alms-giving (zakat), and pilgrimage (hajj)). Next to its simple correlation with the teachings of Islam, Ramadan Fasting is an appropriate measure for the empirical analysis because it varies across both countries and years. This feature results from the definition of fasting in Islam. In detail, Muslims are called to begin the fast with sunrise and to end it by sunset. Thus, the times are country-specific. In addition, the times rotate over the calendar. This feature is related to the fact that Islamic dates follow the Islamic calendar. Unlike the Gregorian calendar that is solar-based and has 365 days per year, the Islamic calendar is moon-based and has only 354 days per year. Due to the shorter period (10–11 days per year), Islamic dates rotate backwards in the Gregorian calendar. It takes 33 years for the Islamic calendar to get back in accord with the Gregorian calendar. Figure 3illustrates the phases for three selected countries. The figure shows the average per-day length of fasting in Chad, Iran, and Indonesia. As discussed in Section 2, the length of fasting is approximated by the average per-day length of daylight in the capital for a 30-days period beginning with the first day of Ramadan. It can be seen that the Ramadan measure varies in a plausible range. The mean value is slightly above 12 hours (since all Muslim-majority countries are slightly above the equator), the curves 9 In Islam, Ramadan is a blessed month ordained by God. During this period, Satan and his minions are said to be locked away in Hell to prevent them from misleading, deceiving, and whispering in the ears of believers. Thus, Muslims consider fasting as an act of deeply personal worship seeking a raised level of closeness to God. Like in other religions, this is done by putting more effort into studying and following the teachings of the religion. Islamic theology supports this behavior by claiming that acts of good deeds earn multiplied rewards while committing sins during this period is greater and more serious than sins at other times. The effectiveness of such claims is documented in a 2013 PEW survey conducted in 39 Muslim countries where a median of 93% of Muslims indicate that they follow the practices of Ramadan. The report is available at www.pewresearch.org. FIGURE 3 The figure shows the development of the average per-day length of Ramadan Fasting for three selected countries. Fasting lengths are measured by the average per-day hours between sunrise and sunset times for a 30-day period beginning with the month of Ramadan in a given year and they apply only to the capital of an individual country. See Section 2for more information. 598 KHODAVERDIAN interact when Ramadan overlaps with the beginning of spring or autumn (capturing day and night equinox), the amplitudes correspond to the distance of the capitals to the equator, and the curves run contrary for capitals above vs below the equator. In our dataset, there are only two Muslim-majority countries with capitals below the equator: Comoros (Moroni) and Indonesia (Jakarta). Even the 33-year cycle is clearly visible. For example, for the countries above the equator, the maximum length of fasting is reached in 1950, 1983, and 2016 (suggestive in the figure since our dataset ends in 2015). When employing the Ramadan measure, we claim that the intensity of the teachings of Islam varies with the length of fasting in a given country. To better understand this claim, it is helpful to specify the situation during Ramadan in the Muslim-majority countries. In longer fasting periods, Muslims have few opportunities other than religion. In some of the countries, restaurants close for the day hours or they run on reduced capacity serving only the demand that comes from those groups that are exempt from the fast (children, elderly, ill people, pregnant women, breastfeeding mothers, and travelers). Similarly, companies with physically demanding work reduce their working hours since workers are exhausted by the fast. In contrast to that, Islamic entities attract the people by offering legitimate alternatives throughout the month. For example, mosques offer gatherings where Muslims can discuss their affairs with other believers (shura). In case of ambiguity, the advice of religious scholars can be sought (ijma). Alternatively, Muslims can join scholarly speeches where issues of modern times are interpreted in the context of Islamic law (ijtihad). The opportunities are utilized by the Muslims spending additional time (i.e. beyond the five mandatory prayers per day) studying and sharing the teachings of the Quran (e.g. Afifi, 1997). To keep the people engaged, activities are extended to the evening. At sunset, prayers are carried out with the community (ummah) and thereafter (free) food is offered to the public to break the fast (iftar). The consequences of the increased engagement are also present in the data. Campante and Yanagizawa-Drott (2015) show that longer fasting hours have a negative effect on economic growth but a positive effect on happiness and life satisfaction of the Muslims. An increased spiritual orientation of the Muslims during this period is also reported by Seyyed et al. (2005) and Bialkowski et al. (2012). It is worth highlighting that the Ramadan measure includes exogenous variation. As explained above, Ramadan is determined by nature. This makes the measure exogenous to all other variables that play a role for the relationship between Islam and democracy. This feature is valuable for empirical work because it enables the estimation of the causal effect that runs from Islam to democracy. Albeit, Ramadan is a meaningful practice only in the Muslim countries. In countries where Islam plays a minor role, that practice has no effect on outcomes (see e.g. Campante and Yanagizawa-Drott (2015)). Thus, we focus on the Muslim-majority countries when applying the Ramadan measure. 5.2 |Estimation Table 8shows the effect of Ramadan Fasting on the measures for democracy in the Muslim-majority countries. The measures for democracy are named in the Panel-head. We present two types of specifications including one where the outcome variable is the level of the democracy index and one where the outcome variable is the change in the democracy index. The top row specifies the outcome variable. The second-top row names the model specification and the third-top row clarifies the empirical model. Columns (1) and (2) represent a basic specification where we estimate the effect of (log) Ramadan on democracy. The specifications show that longer fasting hours have a positive effect on democracy. The inclusion of the first lag of democracy does not matter for the pattern. Moreover, it holds true with both measures for democracy. Even though this specification is insightful, it assumes that Ramadan Fasting is equally important in all the Muslim-majority countries irrespective of whether Muslims represent 75% or 100% of the total population. In columns (3) and (4), we address this issue by weighting the Ramadan measure by the Muslim share (displayed as Muslim in the table to save space). This approach helps identifying where the significance comes from in the basic cases. When democracy is measured by the Political Rights index (Panel A), the static model in column (3) produces insignificance estimates. This pattern suggests that neither Ramadan nor Ramadan fasting performed by the Muslims has an individual effect KHODAVERDIAN 599 TABLE 8 Islam and democracy: ramadan fasting as alternative measure Dependent variable: democracy Panel A: Political Rights index (normalized) Outcome variable Index level Index change Model specification Basic Weighted Weighted & controls Max variation Empirical model Static Dynamic Static Dynamic Static Dynamic Static Dynamic (1) (2) (3) (4) (5) (6) (7) (8) Democracy t-1 0.304*** (0.0993) 0.304*** (0.0996) 0.304*** (0.103) (log) Ramadan t-1 0.571** (0.217) 0.634** (0.234) 0.404 (0.244) 0.615** (0.268) 0.482** (0.203) 0.690** (0.261) (log) Ramadan t-1 * Muslim t-1 0.185 (0.220) 0.0190 (0.170) 0.158 (0.185) 0.0146 (0.162) Initial democracy (;19811985) 0.797*** (0.184) Change in average Fasting length (;19811985 to ;19982002) 0.00857 (0.0237) 0.0170 (0.0196) Country fixed effects Yes Yes Yes Yes Yes Yes No No Year fixed effects Yes Yes Yes Yes Yes Yes No No Observations 281 248 281 248 254 224 39 39 R-squared 0.073 0.170 0.078 0.170 0.083 0.172 0.003 0.360 Countries in sample 36 36 36 36 33 33 39 39 600 KHODAVERDIAN TABLE 8 (Continued) Panel B: Polity 2 index (normalized) (1) (2) (3) (4) (5) (6) (7) (8) Democracy t-1 0.490*** (0.113) 0.475*** (0.126) 0.500*** (0.142) (log) Ramadan t-1 0.305* (0.175) 0.267* (0.150) 0.0836 (0.232) 0.0240 (0.192) 0.0452 (0.161) 0.00844 (0.133) (log) Ramadan t-1 * Muslim t-1 0.436** (0.202) 0.318* (0.164) 0.545*** (0.167) 0.374** (0.154) Initial democracy (;19811985) 0.539* (0.286) Change in average Fasting length (;19811985 to ;19982002) 0.00663 (0.0260) 0.0240 (0.0263) Country fixed effects Yes Yes Yes Yes Yes Yes No No Year fixed effects Yes Yes Yes Yes Yes Yes No No Observations 341 316 341 316 276 261 35 35 R-squared 0.266 0.435 0.294 0.447 0.324 0.486 0.002 0.171 Countries in sample 35 35 35 35 32 32 35 35 Notes: The table shows OLS estimates for the effect of Ramadan Fasting (average per-day hours) on two measures for democracy in countries where Muslims represent ≥75% of the total population. The Panels vary by the measure for democracy. The top row informs about the nature of the outcome variable and the second-top row specifies the model specification. The third-top row names the empirical model. In columns (3)–(4), the first lags of both (log) GDP per-capita and (log) life expectancy are included (not shown in the table). The regressions in columns (1)–(6) are based on a five-year data frequency. Consequently, the subscript t1 refers to the value five years ago. In columns (7)–(8), the average change in the variables is considered. The periods for the averages are 1981–1985 (longest Ramadan) and 1998–2002 (shortest Ramadan). Standard errors (reported in parentheses) are robust against heteroskedasticity and serial correlation at country-level. Significance levels are. * * * p < 0.01, ** p < 0.05, * p < 0.1. KHODAVERDIAN 601 on democracy. In column (4) when the dynamic model is applied, one of the two estimates is significant. Yet, since this is the estimate for Ramadan Fasting, the specification suggests that Ramadan itself has a positive effect on democracy irrespective of the Muslim share in the Muslim-majority countries. The picture changes entirely in Panel B when the Polity 2 index is employed. Both specifications show that not Ramadan itself but Ramadan performed by the Muslims has a positive effect on democracy. 10 The specifications in columns (1) to (4) assume that Ramadan has a direct effect on democracy. However, this is not necessarily true since it is shown that increased Ramadan hours decrease economic growth and increase subjective well-being in the Muslim countries (see e.g. Campante and Yanagizawa-Drott (2015)). Thus, it is possible that our estimates show the effect that these variables have on democracy rather than the effect that Islam has on democracy. 11 We explore this possibility in columns (5) and (6) where we add the first lags of two control variables including (log) GDP per-capita and (log) life expectancy (not shown in the table to save space). 12 In all cases, the coefficients for the controls turn out insignificant. More importantly, the estimates for Ramadan in Panel A and Muslimweighted Ramadan in Panel B remain significant and positive. Columns (7) and (8) present a different type of analysis. Therein, we consider the effect of a change in the Ramadan measure on the change in the democracy index. Unlike the long-term specifications shown in Table 7, the specifications here consider a change over a half-cycle of Ramadan. The idea is to capture the effect of a maximumchange in the measure for Islam. As shown in Figure 3one maximum in the length of fasting is reached in the year 1983. Given the 33-year cycle of the Islamic calendar, the following minimum length is reached in the years 1999/2000. We consider the year 2000. For each of the two extreme years (1983 and 2000), we calculate the average value for both fasting length and democracy index within a 5-year window. Then, we determine the change as the difference between the last average minus the initial average. The resulting figure includes the maximum change in the measure for Islam and its corresponding effect on democracy. In all specifications, that effect turns out to be insignificant. 13 The specifications in Table 8show that even Ramadan Fasting has no negative within-country effect on democracy. If anything, it has a positive effect. This is an important part of the analysis because Ramadan Fasting includes the core teachings of Islam. Together with the main findings, they draw a clearer picture of the effect of Islam on democracy. 6|CONCLUSION Islam is the world's fastest growing religion. According to data by Pew (2017), Muslims will outnumber Christians by 2075 making Islam the world's largest religion. The consequences of this development go beyond the traditional Muslim countries of the Middle East. Due to globalization and mass migration, Muslim communities exist worldwide. Although most of them already live in democratic countries (e.g. Karatnycky (2002)), there is considerable concern as to whether Islam is compatible with democracy. Our analysis confirms that there is a negative cross-country link between the Muslim share and democracy. However, this pattern is not sufficient to blame Islam for the democracy 10 An alternative to weighting the Ramadan measure with the Muslim share is to estimate the effect of Ramadan on democracy only for countries where the Muslim share is ≥95% of the total population. This sample excludes noise in the data since it drops out those countries where Islam's supreme position is not always guaranteed. Depending on the measure for democracy, this specification shows that Ramadan's effect is either positive (Political Rights index) or insignificant (Polity 2 index). In no case though did we find evidence for a negative effect (not shown here but available). 11 As discussed in Footnote 7, income is shown to have no within-country effect on democracy. Yet, Cervellati et al. (2014) discover that this conclusion is driven by heterogeneity related to the colonial history. They show that the effect is negative for former colonies, but positive for non-colonies. Thus, depending on that feature, income has an effect on democracy. When controlling for income though that mechanism is cancelled. 12 We do not have appropriate proxies for subjective well-being. However, it is reasonable assuming that this figure is correlated with GDP per-capita. Moreover, the inclusion of life expectancy as a control variable is justified because the literature documents that Ramadan Fasting has an effect on health (see e.g. Bialkowski et al. (2012) for a discussion). 13 The pattern is the same when we exclude Comoros and Indonesia where the development of Ramadan is opposite since the (capitals of the) countries are located below the equator (not shown here but available upon request). 602 KHODAVERDIAN deficit. As we show, the lower level in democracy is due to fixed differences across the countries. When the differences are considered, there is no evidence for a negative effect of Islam on democracy. This is true irrespective of whether Islam is measured by the scope of the believers or by its core teachings. It also applies to Muslim-majority and Muslim-minority countries alike. If anything, Islam has a positive within-country effect on democracy. Despite extensive analysis, it remains open which of the fixed country characteristics led to the lower initial level in democracy in the Muslim countries. The literature offers two prominent theories including geography and institutions. In Appendix Table A.3, we consider proxies for both theories. Yet, neither of them removes the negative cross-country link. This leaves us with two possibilities: Either Islam did have a negative influence on democracy but the effect has cancelled in modern times or the relationship between Islam and democracy is even more complicated. There is some evidence that speaks against the first idea. For example, Chaney et al. (2012) point out that the democracy deficit is persistent in those countries that were conquered by the Arab armies following the death of Prophet Muhammad in 632 CE. Countries that implemented Islamic law later have developed very similar to the rest of the developing world. Instead, the authors suggest that the institutional framework in the initial Muslim countries is prone to autocratic leadership. If so, the question is whether Islam has promoted weak institutions and thus is responsible for the lower initial level? Again, evidence shows differently. Until 500 years ago, most societies were non-democratic. This pattern changed during the period of European colonization. Depending on existing levels of wealth, the colonizers have established either extractive or inclusive institutions. Countries with extractive institutions perform weaker (e.g. Cervellati et al., 2014). The fact that Islam was irrelevant for the decision-making of the Europeans is shown in Table 8B by Acemoglu et al. (2008). We prefer the second alternative (the relationship between Islam and democracy is even more complicated). The idea of modern democracy began in the US and it was brought to Europe in the second half of the 19th century. In the beginning of the 20th century, it spread to some countries in Latin America. Since all the countries have a Christian background, leaders in the non-Christian countries (whether Islamic or not) began to form an opposition. To be effective, they focused on factors that are commonly shared by the people, e.g. culture, creating a national identity. In the countries of the Middle East, Islam served this purpose. On the one hand, the religion has no generally accepted representative who can speak for all the Muslims in the world. This leaves room for adjustments. On the other hand, Islam has a long history and a reputation that the people trust. This setting is well-suited to direct the people (e.g. Rohac (2013)). This kept democracy out of the Muslim world leading to a lower initial level. 14 Still, it is not an Islamic problem. This is shown in Appendix Table A.3 where the negative cross-country link applies to all countries (i.e. all latitudes) even after controlling for the Muslim share. The effectiveness of such tools (national identity or religion) is changing over time though. New media tools became available that enabled the exchange of information outside of the state control. Thus, the idea of democracy is spreading across the world. As shown in Figure 1, the Muslim countries are also affected by that development. It is possible to support the development. Policy-makers can emphasize that neither Muslims nor the teachings of Islam (as a religion) poses a threat to democracy. This reduces prejudices in the population and it offers the opportunity to get to know each other working out the values that are important to all the people. ACKNOWLEDGMENTS I thank the editor David Stadelmann and two anonymous referees for very useful comments. I am also thankful to Thomas Apolte, Michael Berlemann, Andreas Fuchs, Dierk Herzer, Hamza Safouane, Henning Schäckelhoff, Rainer Thiele, and Stefan Voigt for comments and suggestions. Sabrina Wendt has offered excellent support. Seminar participants at the North American Summer Meeting of the Econometric Society, World Interdisciplinary Network for Institutional Research (WINIR), Southern Economic Association, HSU-IfW (Kiel Institute for the World Economy) Workshop in Development and Environmental Economics, University of Münster, and University of Hamburg have 14 Natural resources add to the issue since they represent a source of income that is independent of the approval of the people (e.g. Ross (2001)). Yet, it is important nothing that the democracy deficit predates that income source. Thus, oil and gas cannot explain the origin of the lower initial level of democracy in the Muslim countries. KHODAVERDIAN 603