Seimo rinkimų rezultatų prognozavimas : valdančiųjų partijų balsų modelis

LTStraipsnyje siekiama pasiūlyti pirmąjį Lietuvoje Seimo rinkimų rezultatų prognozavimo modelį, paremtą statistine analize. Pirmoje dalyje apžvelgiami svarbiausi šios akademinės tradicijos aspektai ir pateikiami kriterijai, pagal kuriuos vertinamos prognozių funkcijos. Antroje dalyje suformuojamos dvi svarbiausios modelio prielaidos: pirma, kaip priklausomas kintamasis analizėje traktuojamas atskirų valdžios partijų pasirodymas Seimo rinkimuose (taip išvengiant mažos imties problemos) ir, antra, prognozuojama pagal jų balsus savivaldos rinkimuose (nepriklausomas kintamasis). Remiantis šiomis prielaidomis, trečioje dalyje pateikiamas tiesinės regresijos modelis ir išvedama prognozavimo formulė. Taip pat aptariami galimi būdai ją patobulinti, įvedant papildomus kintamuosius (svarbių politinių aplinkybių veiksnys) ir pasiūlomas antras modelio variantas. Pagal abu modelius pateikiamos konkrečios prognozės trims valdžios partijoms (TS-LKD, LRLS ir LiCS): kiek jos gaus balsų (išreikštų procentais) daugiamandatėje apygardoje 2012 m. Seimo rinkimuose. [Iš leidinio]Reikšminiai žodžiai: Prognozių modelis; Rinkimo rezultatų prognozavimas; Rinkimų rezultatų prognozavimas; Savivaldos rinkimai; Seimo rinkimai; Valdžios partija; Election to Seimas; Elections to Seimas; Forecasting model; Forecasting of election results; Government party; Municipal elections.

ENElectoral forecasting has been quite heavily investigated in the Western countries (especially USA, Great Britain and France). However, models that strive to forecast the election results in new democracies are quite rare. This article aims to present first forecasting model for the elections to Lithuania's parliament Seimas. Since sample of available democratic elections is too small for the statistical modeling, treating the votes of separate government parties as a dependent variable is suggested: assumption is made that it is possible to forecast the votes of different parties at the same elections, while using one model. Though party system fragmentation is usually treated as a negative factor for the electoral forecasting, in case of Lithuania it enables building of a statistical model with at least 9 cases (sample used in the article). Another important assumption of the model is that performance of parties in the parliamentary elections is linked to their performance in the preceding local elections. Resting on these assumptions, model that strives to predict the shares of votes of separate government parties (parties that spend major part of the term between elections being in the national government) is constructed. Using just one independent variable (share of votes at the municipal elections), model explains 0.913 percent of variance in the dependent variable (adjusted R-squared) and has a Standard Error of Estimate (SEE) equal to 1.782.These characteristics show quite satisfactory accuracy of the model; it also satisfies all three requirements for forecasting functions (parsimony, reproducibility, lead) quite well. Ways to improve the suggested model are discussed: one of the possible improvements is inclusion of dummy variables measuring political circumstances that could affect the votes of the parties. When two dummy variables (negative and positive political circumstances/events) are added to the model, R-squared and SEE improve significantly: however, such model is vulnerable to criticism because of the subjectivity and poor lead. Resting on both models, forecasts of the votes (percent) for the three government parties in the forthcoming 2012 elections to Seimas are made. [From the publication]