The turnout figures for all Dutch municipalities in 13 national parliament elections held from 1971 to 2010 were obtained from the Dutch Electoral Council. The municipalities were classified according to the 2010 codes issued by Statistics Netherlands. During the period in question the number of municipalities increased from 412 to 425 as a result of land reclamation. Municipality turnout, defined as the percentage of eligible voters in a municipality who actually voted in the election, varied from 56.5 to 100%, with a mean of 83.3% (SD 5.8).
The weather statistics of 17 to 35 meteorological stations and 222 to 235 precipitation stations were obtained from the Royal Dutch Meteorological Institute, with the number of stations trending up over time by the expansion of weather recording. The measurements include election-day mean temperature in degrees Celsius, fraction of maximum possible sunshine duration in percentages, and rainfall amount in millimeters. The temperature and sunshine data were obtained from the nearest meteorological station. The station closest to the municipality was determined with the Haversine formula, rendering the distance between municipality and station from their longitudes and latitudes. The mean distance was 17.7 km (SD 9.8). A comparison of recorded data at neighboring stations indicated that a greater density of weather stations adds little to the accuracy of the local temperature and sunshine data. The daily mean outside temperature ranged from 1.7 to 19.9°C, with a mean of 12.0°C (SD 4.4). The fraction of maximum possible sunshine duration ranged from 0 to 93%, with a mean of 42.3% (SD 29.3). As local variability is more of an issue with precipitation, the amount of rainfall was obtained from the nearest precipitation station. The mean Haversine distance between municipality and precipitation station was 4.4 km (SD 2.3). Local rainfall on election day ranged from 0 to 43.9 mm (mean 2.6, SD 4.0).
Municipality turnout rates exhibit both continuity across elections and uniqueness. Turnout rates for successive elections are strongly related as a result of routing voting and voter abstention. To account for this temporal continuity the model includes the municipality turnout rates in the two preceding elections. The turnout rates for the elections prior to 1971 were additionally coded to obtain complete data for the first two elections. Because contemporary issues are unique to each election, election dummy variables were entered that allow the election intercepts to vary. Also, the national parliament elections held from 1971 to 2010 comprise eight regularly scheduled elections and five early elections. The actual election dates are listed in Table 1.
If the Dutch government remains in office for the complete term, elections for national parliament are held once in every four years and they are typically scheduled in May, when the weather is at its most pleasant. However, voters have also been called to the poll for an early election in late summer, autumn and winter, following the untimely fall of the government. To examine if the weather effects on turnout vary by season, interaction terms of the weather variables and the seasonal photoperiod were included in the model. Photoperiod was obtained as the time from sunrise to sunset in minutes, using the geographical centre of the Netherlands as reference point. Election-day day length ranged from 611 to 1,117 min (mean 938, SD 181). The variable included in the analysis is photoperiod divided by 60.
The municipality longitudes and latitudes (in degrees, decimal degrees) were included to account for regional variations across the country in factors that may affect turnout, such as political apathy, religious affiliation and other correlates of political participation. Finally, the municipality voting-age population density was used as demographic control. The variable included in the analysis is the natural logarithm of the number of eligible voters per municipality square kilometer.
The data were analyzed using three-level hierarchical models—both linear and logistic—with voters at level one, nested within municipality-by-election at level two and municipality at level three. As the parameter estimates of the two models obtained identical effect signs and near equivalent p-values, we opted for the presentation of the results of the linear model as they are easier to interpret.