We have already discussed the problems of constructing a
fundamental model based on estimates from historical data due to the temporal instability
caused by the participation of new parties (Podemos and C’s) in the 2015
Congressional election. To avoid this temporal instability we need to
concentrate in recent information and cross sectional data.
Following this strategy we have constructed a fundamental model for electoral sentiment at provincial level using "deep" pre-electoral survey data (see below), Bayesian hierarchical models and post-stratification techniques. The post-stratification takes into account gender, age groups, education level, professional status and population density, and it is based on census data. Finally, using post-electoral survey data we estimate predict participation.
Following this strategy we have constructed a fundamental model for electoral sentiment at provincial level using "deep" pre-electoral survey data (see below), Bayesian hierarchical models and post-stratification techniques. The post-stratification takes into account gender, age groups, education level, professional status and population density, and it is based on census data. Finally, using post-electoral survey data we estimate predict participation.
To train the data we use recent samples that included the
option of the new parties. Training data includes CIS pre-electoral poll for
the 2014 European elections and pre-electoral polls for 2015 autonomic
elections in the CCAA where elections took place.
Figure 1 shows the performance of our method with respect to the actual outcome of the European elections. The results are significantly better than the CIS estimates and show a reasonable accuracy.
Figure 1.
Figure 1 shows the performance of our method with respect to the actual outcome of the European elections. The results are significantly better than the CIS estimates and show a reasonable accuracy.
Figure 1.
