Fig 1.
A schematic diagram of the process flow in this paper, showing the different types of models, including their inputs and outputs.
Fig 2.
A summary of the different models proposed in this paper.
Table 1.
Exploring space of outcomes (Seat shares) for different vote shares using DPM and PCM models with different concentration parameter settings.
Fig 3.
Winner maps of two elections over 100 seats, simulated by DPM (left) and GDPM (right).
The vote share is same in both cases and also the seat share
.
Table 2.
Social identity model in two scenarios ,
, for
and
. Above: individual preference, Below: local influence variants of SIM.
Fig 4.
Swing from previous election (left) to next election (right) under Dirichlet Swing Model, with parameter (top right) and
(bottom right).
Fig 5.
Vote Share Swing vs Seat Share Swing for different parties under different different swing models and parameters.
Left panel: Districtwise Swing Matrix Model (DSMM), Right Panel: Districtwise Swing Model (DSM). Circle ’o’ denotes Party 1, Plus ’ + ’ denotes Party 2, Star ’*’ denotes Party 3. Each color (blue, red, green, black) indicates a different parameter setting.
Table 3.
Details of a few past elections from different states in India.
Table 4.
Simulation of seat shares in 4 Indian elections by DPM (middle) and PCM (below) under optimal and default parameter settings.
The actual seat shares in these elections are shown in the upper part of the table. Simulated results in case of optimal parameter settings match the actual results in all cases.
Table 5.
Comparing spatial correlations between vote shares (VSC) and winners (WSC) in actual elections and elections simulated by DPM, GDPM, PCM, GPCM.
The simulated results that are closest to the actual results are highlighted.
Table 6.
Comparing swings in vote and seat shares across successive elections in reality and as simulated by DSM and DSMM.
Upper 5 rows are for swings across GJ17 and GJ22, while the bottom rows are for swings across WB19 and WB21. The results of GJ17 are , and those of WB19 are
.
Fig 6.
Comparing the error in predicted Vote Share (X-axis) and prediced Seat Share Swing (Y-axis) for different parties under different swing models and parameters.
Left panel: Predictions of West Bengal Elections 2021 (WB21) based on West Bengal Elections 2019 (WB19). Right panel: Predictions of Gujarat Elections 2022 (GJ22) based on Gujarat Elections 2017 (GJ17). Circle ’o’ denotes Party 1, Plus ’ + ’ denotes Party 2, Star ’*’ denotes Party 3. Blue and Red colors indicate predictions by DSM model, while Green and Black colors indicate predictions by DSMM model.
Table 7.
Accuracy of seat projections due to variation of on some Indian elections.
We report the mean Manhattan Distance in each case, while the probability of accurate projection is shown in brackets.
Fig 7.
Change in seat projection errors (Manhattan Distance) due to variation of on elections simulated by DPM (upper part: DPM-6, lower part: DPM-7).
Fig 8.
Change in seat projection errors (Manhattan Distance) due to variation of on elections simulated by PCM (upper part: PCM-6, lower part: PCM-9).
Fig 9.
Change in seat projection errors (Manhattan Distance) due to variation of on elections simulated by SIM (upper part: SIM-1, lower part: SIM-2).
Table 8.
Change in seat projections due to variation of on Indian elections for direct and swing-based projections.