Beliefs, economics, and spatial regimes in voting behavior: the Turkish case, 2007–2018

The paper investigates why the AKP has maintained substantial electoral support nationwide despite economic slowdowns and political turmoil after its initial growth-led success (average 7.3% growth in 2002–2007; slower growth thereafter). The research asks whether economic performance (EVT) or structural cleavages (Center–Periphery) best explain voting behavior for the AKP versus Pro-Kurdish parties, and whether spatial heterogeneity shapes these relationships. The study aims to disentangle pocketbook and sociotropic components of EVT from C–P drivers (religion, ethnicity, space), measure their relative importance, and visualize spatial patterns and regimes of voting across Turkish provinces from 2007 to 2018.

Economic Voting Theory posits voters reward or punish incumbents based on economic performance via two mechanisms. Pocketbook (egotropic) voting evaluates personal economic conditions (e.g., unemployment, inflation, income growth), assuming voter rationality and sophistication. Sociotropic voting emphasizes national economic performance (“the country’s pocketbook”), allowing support even if personal conditions do not improve. Ideology and partisanship can interact with perceived economic outcomes, potentially biasing EVT estimates. Beyond EVT, redistributive politics and structural cleavages such as religion, ethnicity, age, and gender shape partisanship and vote choice. The Center–Periphery (C–P) framework (Lipset & Rokkan) explains how territorial-cultural conflicts (language, religion, morality) and peripheral identities condition party systems and alignments, often outweighing purely economic bargaining. In Turkey, long-standing center–periphery tensions include secular civil–military elites versus religious–conservative rural constituencies, and the Kurdish population’s demand for recognition. Prior Turkish studies variously emphasize C–P cleavages (conservative–secular, Turkish–Kurdish, east–west migration) and EVT effects, but the spatial dimension of these mechanisms has been underexplored. This study addresses that gap by measuring and visualizing spatial aspects of EVT and C–P in Turkish elections.

Design: Comparative empirical analysis integrating EVT and C–P mechanisms using provincial panel data (81 provinces) across five general elections: 2007, 2011, June 2015, November 2015, and 2018. T = 5. Data sources: TurkStat (TUIK), Directorate of Religious Affairs (Diyanet), and Supreme Election Council (YSK). An online Dataverse repository provides the dataset. Variables: - Outcomes: Provincial vote shares for (i) AKP and (ii) Pro-Kurdish parties (various labels over time). - EVT (economic) predictors: National GDP growth (sociotropic), local per capita GDP growth, local unemployment rate, local inflation (pocketbook). To address voter myopia, national growth, local growth, and local inflation are time-weighted over the four quarters preceding each election (following Akarca & Tansel 2006). - C–P predictors: Religion (provincial share of Qur’an course attendees; Diyanet), ethnicity (KurdPop: provincial percentage Kurdish). Mosque counts were avoided due to legal/institutional biases from the 1924 Village Law; Qur’an course attendance better captures religiosity and conservative orientation, including adult participation and the post-2012 surge after age restrictions were removed. - Controls: MedianAge, higher education rate (Hedu), Depreciation (incumbent vote-share depreciation: interaction of lagged vote share with years in power), StrategicVoting (first lag of the party’s vote share), and an election-period dummy d18 capturing 2016 coup attempt and 2017 referendum effects prior to 2018 election. Modeling strategy: 1) Panel regression with province fixed effects (α_i) and idiosyncratic errors μ_it: Party_it = α_i + β1 NationalGrowth_it + β2 LocalGrowth_it + β3 LocalUnemp_it + β4 Religion_it + β5 KurdPop_it + β6 MedianAge_it + β7 Hedu_it + β8 d18_it + β9 LocalInfl_it + β10 Depreciation_it + β11 StrategicVoting_it + μ_it. White (heteroskedasticity-robust) standard errors are used. 2) Spatial analysis: Local Indicators of Spatial Association (LISA) maps (queen contiguity) identify spatial clusters and outliers in party vote shares (2002–2018 for visualization of long-run clustering). Spatial regimes are then defined by ethnicity: Regime 1 = provinces with Kurdish population share ≥15% (eastern/southeastern); Regime 0 = others. Chow tests assess structural breaks globally and by regressor between regimes. 3) Spatial diagnostics (LM-lag and LM-error; robust versions) guide selection of spatial process per regime: AKP models indicate SEM in Regime 0 and SAR in Regime 1; Pro-Kurdish models indicate SAR in Regime 0 and no spatial model in Regime 1. Spatial decisions suggest no spillovers across regimes. Descriptive coverage: 405 province–election observations. Key descriptive means (N=405): AKP vote share 47.0% (SD 14.1), Pro-Kurdish 11.7% (SD 20.4), NationalGrowth 5.47% (SD 1.94), LocalGrowth 5.54% (SD 4.09), LocalUnemp 9.54% (SD 0.87), Religion (Qur’an course attendance share) mean around 1.24% with variation over time and space. Ethnicity measurement: Due to absence of official local ethnic data, Kurdish shares are derived from Mutlu (1996) using 1965 mother-tongue, fertility differentials (TDHS), migration, and return flows, spatially interpolated from 67 to 81 provinces to align with current boundaries.

Nationwide and non-spatial panel results (Table 2) and spatial-regime analyses (Tables 3–4) show: - EVT mechanisms: - Sociotropic voting dominates pocketbook voting for the AKP. NationalGrowth is positive and significant for AKP; local economic variables (LocalGrowth, LocalInfl, LocalUnemp) are mostly insignificant nationwide. For example, in a representative AKP model (regression 3), NationalGrowth ≈ +2.034 while Religion ≈ +7.175. - In spatial regimes, sociotropic voting (NationalGrowth) remains positive and significant in both Regime 0 and Regime 1 (Table 3: +2.590 and +2.204, respectively; no structural break by Chow). Pocketbook effects appear only in Regime 0 via LocalUnemp (−1.109, p<0.01). In Regime 1, LocalGrowth is significantly negative (−0.704), counter to standard pocketbook expectations. - For Pro-Kurdish parties, NationalGrowth is significantly negative in Regime 0 (−0.462), indicating that when the national economy performs better, some votes shift away from Pro-Kurdish parties (consistent with AKP gains). No significant sociotropic effect appears in Regime 1. - C–P mechanisms: - Religion strongly predicts AKP support, outperforming economic variables. In regime models, Religion is positive and significant in both regimes, with a larger coefficient in Regime 0 (+9.299 vs +4.742; Chow p=0.019 indicates a structural break). For Pro-Kurdish parties, Religion is negative and significant in both regimes (−0.915 in Regime 0; −3.056 in Regime 1 at 10%). - Kurdish population share (KurdPop) negatively correlates with AKP vote share and positively with Pro-Kurdish vote share. In AKP regime models, KurdPop is significantly negative in Regime 1 (−0.301) and not significant in Regime 0; for Pro-Kurdish parties it is positive and significant in both regimes (~+0.52 to +0.53). - Education and age: Higher education (Hedu) is negatively associated with AKP vote share (Regime 0: −0.456; Regime 1: −2.656) and positively with Pro-Kurdish vote share (Regime 0: +0.347; Regime 1: +4.276). MedianAge is negatively related to AKP support in Regime 0 (−1.755) but not significant in Regime 1; for Pro-Kurdish parties, MedianAge is significantly negative in Regime 1 (−3.509), implying younger populations there are more supportive. - Spatial structure and regimes: - LISA maps show stable high–high AKP clusters in central/central-east Anatolia and the northeast Black Sea, and low–low clusters in Kurdish regions; Pro-Kurdish high–high clusters persist in the east/southeast, aligning with Kurdish population concentrations. - Chow tests indicate significant global structural breaks between regimes for both parties (AKP GlobalChow ≈ 71.0; Pro-Kurdish ≈ 93.0; p<0.001), with variable-specific breaks (e.g., LocalGrowth, LocalUnemp, Religion, MedianAge, Hedu, Depreciation for AKP). Spatial diagnostics suggest distinct spatial processes within regimes and no spillovers across regimes. - Model controls: Depreciation behaves unexpectedly for AKP in some nationwide specifications (positive), while StrategicVoting is generally negative and significant for both parties when included, and can mask other effects; models without StrategicVoting yield consistent substantive patterns. Overall: C–P factors (religion, ethnicity, space) have stronger and more consistent effects than EVT variables. Where EVT operates, sociotropic effects are stronger and more pervasive than pocketbook effects.

The findings explain sustained AKP electoral support despite economic slowdowns by showing that structural Center–Periphery cleavages—especially religious conservatism and ethnic composition—dominate over economic conditions in shaping vote choice. Sociotropic evaluations of national performance bolster AKP support across regimes, while pocketbook responses are limited to non-Kurdish majority provinces and are smaller in magnitude than religious effects. Religious mobilization policies (e.g., easing access to Qur’an courses) yield higher electoral returns than improvements in local economic conditions and even exceed the effect of national growth. For Pro-Kurdish parties, support hinges on ethnic identity (KurdPop) and is undermined where religious conservatism is stronger; national economic upswings reduce their vote share outside the Kurdish heartland. Spatial analyses reveal distinct, non-interacting voting regimes aligned with ethnic geography, reinforcing polarization and limiting cross-regime persuasion. Together, these results indicate that while economic performance matters (mostly sociotropically), enduring peripheral cleavages provide the primary foundation of electoral behavior, accounting for the resilience of AKP support during periods of weaker economic performance.

The study shows that voting for the AKP and Pro-Kurdish parties is driven by a combination of EVT and C–P dynamics, with C–P factors exerting the larger influence. Sociotropic voting is common nationwide and across spatial regimes, whereas pocketbook effects are confined to non-Kurdish majority provinces and are weaker. Religious conservatism powerfully increases AKP support, surpassing the effect of national growth, while Kurdish population share depresses AKP votes and boosts Pro-Kurdish support. Spatial regime tests confirm structurally different processes across Kurdish-majority and other regions without spillovers. These insights suggest that for incumbents like the AKP, strategies appealing to religious identity can be more electorally effective than economic performance alone, helping sustain dominance under economic headwinds. Future research could deepen causal identification of religious policy effects, incorporate finer-grained local economic indicators, and extend spatial regime analyses to municipal elections and post-2018 contests.

- Ethnicity data are not officially available at the local level; Kurdish population shares are inferred from Mutlu (1996) using 1965 mother-tongue, fertility differentials, and migration, then spatially interpolated from 67 to 81 provinces. This introduces measurement uncertainty and temporal lag relative to 2007–2018 elections. - Religiosity is proxied by Qur’an course attendance, which increased sharply after the 2012 removal of age restrictions and was encouraged by the AKP. While informative, this proxy may capture policy-driven participation dynamics alongside underlying religiosity. - Mosque counts, a common religiosity proxy, were avoided due to legal/institutional bias (1924 Village Law), but the chosen proxy also has limitations. - Inclusion of a lagged dependent variable (StrategicVoting) can inflate model fit and attenuate other coefficients; some specifications exhibit counterintuitive signs (e.g., religion for AKP) when the lag is included, suggesting sensitivity to dynamic specification. - Spatial diagnostics indicate distinct processes within regimes, but regime assignment via a 15% Kurdish-share threshold, while empirically grounded, is a discretization that may mask within-regime heterogeneity.