Traditional way of estimating parties’ chances

• In a riding or at the national level

How is it Calculated?

What Has Been Done With This Approach?

• Dufresne et al. (2018): Sur le potentiel de croissance du PQ (On the Potential for Growth of the PQ)
• Déry et al. (2022):
• Unpublished Cadieux et al. (2023): On Volatility and Parties’ Potential for Growth

{Montrer des graphs intéressants de ces notes de recherche}

Theoretical Benefits of Such Measure

• More suitable to multiparty systems
• Adding layer of choices → reduction of prediction error
• More nuance to deal with undecided voters
• Alleviate uncertainty associated with attitude/behaviour discrepancies
• Makes a difference between unreachable and reachable electors → potential for growth
• Makes a difference between a solid and a fragile voter → vote solidity

Some individual validity

{What does it mean to be a 3? A -2? A 6? Mettre les probabilités prédites sur l’intention de vote binaire}

Descriptive: RCI distribution by party

{Distribution de l’IRC des partis prov et féd}

Case studies

• Canada Federal Election, 2021
• Quebec Provincial Election, 2022

Identifying electoral swings

• Estimate the expectation in a riding using vote expectations (wisdom of the crowds)
• Compare with actual elections results

Descriptive graphs: electoral swings

Estimating the RCI at the riding-level

• Why? Only way to compare actual election results to survey data
• Step 1: compute regression models to predict the RCI using survey data
• Step 2: predict those models on riding-level census data
• Step 3: aggregate the RCI and its uncertainty using segments’ weights in the riding

Step 1: compute regression models to predict the RCI using survey data

• Vote solidity: lm(rci ~ age * langue * region * male) but only for a party’s current electoral base
• Potential for growth: lm(rci ~ age * langue * region * male) but only for respondents who do not currently vote for the party
• Vote intention: multinom(party ~ age * langue * granular + male)

Step 2: predict those models on riding-level census data

Step 3: aggregate the RCI and its uncertainty using segments’ weights in the riding

Output

{Faire un graphique où on ordonne les ridings pour la CAQ avec les 3 modèles (potgrowth vote solidity et vote intent) sur le graph}

{Faire un graphique qui résume pour une circonscription avec les 5 partis ou pour 2-3 segments dans une circonscription}

The End of the Line?

Weak Signal, Yet Still Worth Exploring

• Preliminary weak findings: caution against premature dismissal of the concept
• Solid theoretical foundation, derived from logical intuitions
• Only a handful of operationalisations have been explored as of yet
  • Broad range of research possibilities

Improving its Conceptualization

• Catalysts
  • Electoral campaign
  • Salient events
  • Parties’ efforts to reach segments
• Position of parties and respondents on salient issues
  • Parties’ room to maneuver
• Political sophistication

Operationalisation of the Concepts

• Challenge: need to start with simplistic measures before going into more complicated operationalisations
• Catalysts

Improving the Models

• Adding independent variables (income, education, etc.)
  • Need more data points to add independent variables → VPL (Vote Compass)
  • (For the synthetic post-stratification table)
• Testing other types of models
  • Random forest
  • Gradient boosting

Survey Experiments: Validate at the Individual Level

• Panel before and after election to follow the variation across the same individuals