Studies on turnout define their dependent variable as the ratio of the number of voters to one of the following: the entire population, the voting-age population, the eligible population, or the number of registered voters. Although the first two are considered often due to the ease with which they can be obtained, obviously the latter two are more proper and will be our choice. These are identical in our case due to the switch in Turkey in 2008 to a system of automatic registration of all citizens eligible to vote without any effort on their part3.
According to Geys (2006), who surveyed 83 aggregate-level studies, in explaining turnout, researchers typically consider socio-economic and demographic variables such as population’s size, density, stability, homogeneity, and the turnout rate in the previous election, political variables such as effective number of parties and the level of electoral competition, and institutional variables such as the election system, registration and voting requirements, and the presence of other elections on the ballot. He notes that among these variables, those on population size, population mobility and political competition appear to be most important, and those on population concentration and homogeneity are not important at all. The findings on variables relating to political fragmentation on the other hand are ambiguous. Although some studies also consider the age and education structures of the population, Geys disregards them on the grounds that interpretations of their parameters carry a potential danger for ecological fallacy. However, Smets and Van Ham (2013), who surveyed 90 individual-level studies, note that education and age are the two most common independent variables in such studies, and the two that are found to be most successful. We believe that as long as one is cautious about deducing individual level relationships from patterns observed in aggregate-level data, controlling for age and education would be beneficial rather than harmful.
For brevity, we will not review all of the papers published since the Geys (2006) survey. These studies considered variables similar to the ones used by earlier studies, as we will do here. However, we will ignore institutional variables relating to election system and type, day of the week and month of the year the election is held, time elapsed since the last election, voting age, whether voting is compulsory or optional, and the ease of registration, as these do not vary in our data. The 2011 election is held on the same day and under the same rules in every province.
The main determinants we consider are the following:
PRIMARY: Percentage of provincial population with at least primary school (5 years of) education.
HIGHER: Percentage of provincial population with at least university education.
UNDER30: Percentage of provincial population between the ages of 20 and 294.
OVER60: Percentage of provincial population over age 60.
URBAN: Percentage of provincial population living in provincial and district capitals.
MP: Number of parliament members elected by the constituency5.
PARTIES: 10,000 divided by the sum of squared vote shares of the AKP, CHP, MHP and BDP6.
NOCOMPETITION: A dummy variable which takes the value of 1 if all parliamentary seats of a province are won by one party, and zero otherwise7
EMIGRANT: Percentage of those born in the province living in another province.
IMMIGRANT: Percentage of provincial population born in another province.
MP × IMMIGRANT/100: Product of MP and IMMIGRANT, divided by 100.
The first eight of the above variables are included in our model to control for socio-economic, demographic, political and institutional characteristics of provinces so that we can estimate the parameters of the last three variables more accurately. Although the results pertaining to the first group of variables are important in their own right, here our main focus will be on the impact of the last three, dealing with migration.
The motivation for considering PRIMARY is that without some minimum level of education, the act of voting alone could be a difficult task, let alone gathering and evaluating information on candidates, parties and issues facing the country and the province. We presume that for most people, primary school education can be taken as that critical level. Although one’s faculty for gathering information and voting rises with increased education, it is likely to be subject to diminishing returns. Furthermore, the opportunity cost of one’s time rises as his/her education rises, especially after graduation from university. The HIGHER variable is added to see if the effect of education on turnout dampens when the province has more people with higher education.
As one ages, one accumulates resources, becomes more experienced, more informed, more settled, and acquires a greater sense of responsibility. These will increase the probability that he/she will vote. On the other hand, the opportunity cost of time increases and health deteriorates as one gets older, which creates disincentives to vote. Consequently, the age-turnout relation is likely to be curvilinear. For that reason, modelers often include in their turnout equations, in addition to age, age-squared.
Individuals born and raised in the same time period are exposed to the same socio-historical events which shape their political socialization. Consequently, political participation may vary between generations too. Bhatti and Hansen (2012) argue that this may be the reason behind the curvilinear relationship observed between age and turnout. For instance, the old may turnout in larger numbers than young individuals not due to age per se, but simply because they belong to a generation with higher turnout levels. It is not possible to separate the age and generation effects from each other in aggregate-level studies examining a single election, such as ours. We have included in our equation UNDER30 and OVER60 variables to measure the combined effects of age and generation.
Individual-level studies find that participation in elections is much higher in rural areas than in urban areas. Voting in urban areas is more cumbersome and the stigma associated with not voting is less as it will be hardly noticed in the anonymity of the city. Villages on the other hand are closely knit societies where each person has intimate knowledge about the activities of others. To capture this, the URBAN variable is considered. However, in Turkey, the urbanization rate is not defined as the proportion of the provincial population living in cities over a certain size but as the proportion residing in provincial and district capitals. As many district capitals are really small towns, only slightly larger than large villages, this definition does not fully reflect the urbanization level of a province. As highly urbanized provinces have large number of parliament members, the MP variable probably captures the urbanization level better. With the MP variable in the equation, the coefficient of URBAN can be interpreted as how the turnout differs between truly urban areas and typical district and provincial capitals. There are other reasons for utilizing MP. It reflects also the population size of the constituency, the complexity of the ballot, and the cost involved in gathering information about the candidates.
A larger effective number of parties (PARTIES) raise the probability that voters can find a party with which they identify, encouraging participation. On the other hand, it also makes it harder for the voters to make up their minds. Also, in a proportional election system, if the share of the top party is relatively high, dispersion of the rest of the votes among many other parties may reduce competition by letting the top party capture a disproportionate number of the seats. The last two reasons may discourage the voters to turn out. Thus, the impact of political fragmentation on turnout is ambiguous.
In parliamentary elections, competition is nationwide. Even when an election in a province is very lop-sided, as long as a voter has a chance to affect the allocation of a seat, he/she will have an incentive to cast a ballot. When a dominant party sweeps all of the seats in a constituency, that chance is extinguished. NOCOMPETITION variable is considered to measure the effect of such an occurrence8.
We now turn to our variables of main interest, EMIGRANT and IMMIGRANT. The coefficient of the former measures the impact of migration at the origin and the latter at the destination. We expect both variables in question to be inversely related to the turnout. There are at least three reasons for this in the case of EMIGRANT. First, people who emigrate are likely to be the ones who are most active politically. Second, remittances sent by these people to their relatives back home reduce the latter’s dependence on the state and thus decrease their incentives to get involved with politics. Third, the ones left behind may be just waiting for their turn to migrate and thus lose interest in local affairs. In the case of IMMIGRANT, if voting is habit-forming as some studies on other countries show, and if the political behavior of the migrants is similar to the behavior of the people in their origin areas – as Pikkov (2011), Akarca and Başlevent (2010) and Akarca and Tansel (2007) show–then we would expect a smaller portion of immigrants to cast a ballot compared to the native born population. As explained above, migrant producing provinces have low turnout rates. Also, provinces with immigrants from all over the country have less sense of community. Furthermore, immigrants are too busy trying to make it in the big city to spare time for political activity. They have less knowledge of candidates and issues at their new locations, and those issues may not be their own. Consequently fewer of them vote. However, an exception has to be made in the case of large urban constituencies where high numbers of immigrants from particular regions of the country are concentrated and where the number of deputies being elected is large. In such provinces, seeing an opportunity to elect one of their own, immigrants will have higher incentives to participate. The interaction term of MP and IMMIGRANT is introduced to capture this.
It will be useful to also consider two other versions of the main model, one which adds the turnout rate in the previous parliamentary election (TURNOUT20007) and another which adds regional dummy variables. Many studies indicate that voting is habit-forming. People who voted in the past are more likely to turnout in future elections. The first extension would take this into account. However, there is a drawback to including a lagged dependent variable in the model. The previous turnout is likely to be effected by the very same variables that current turnout is, as socio-economic, demographic, political and institutional variables change very gradually over time. Thus, at least to some extent, this variable would capture the effects of other independent variables besides habit formation. Consequently it is wise to estimate the model with and without this variable. The second extension would enable us to check if variables included in the model adequately represent regional differences in Turkey and to capture them if they do not. For this purpose, we partitioned the country into four regions (WEST, NORTH, CENTRAL and EAST), shown in Figure 3, based on a finer breakdown by Turkstat9.