Experimental Design
The study uses a six-cell, between-subjects laboratory design. Participants act as candidates competing in repeated pairwise elections. Each participant is assigned to one treatment condition for the entire experiment and does not participate in any other condition.
In each session, 8 participants form one fixed matching group. The session implements a single treatment condition and constitutes the independent unit of observation. Participants complete 30 paid election rounds. The first 10 rounds are designated as learning rounds, while rounds 11–30 constitute the primary analysis period.
At the beginning of each round, the software randomly assigns five participants to Party A and five to Party B. Each Party A candidate is then randomly and anonymously matched with one Party B candidate, producing five simultaneous elections. Party assignments and pairings are redrawn before every round. Participants may therefore encounter the same opponent more than once, but identities are never disclosed.
Within each election, both candidates simultaneously choose a fiscal platform consisting of public-good spending, (g), and private rents, (r). Both variables are selected on an integer grid from 0 to 100 and must satisfy g+r <= 100.
The total fiscal proposal is defined as (t=g+r). Public-good spending benefits the electorate, whereas rents constitute the candidate’s private payoff if elected. Platform choices are rescaled into model units by the software before electoral probabilities and payoffs are calculated.
A computerized electorate evaluates the two submitted platforms according to a fixed treatment-specific rule. This rule converts the difference between the candidates’ platforms into their respective probabilities of winning. The electoral rule remains unchanged throughout a session. Participants receive a qualitative explanation of how the electorate responds to platforms but are not shown its structural parameters or complete mathematical formula.
Before submitting a platform, candidates may use an exact payoff calculator. The calculator allows them to enter hypothetical platforms for themselves and their opponent and displays the corresponding winning probability and expected payoff. Calculator use is voluntary, and all queries are recorded. The submitted-platform calculation and the calculator use the same server-side payoff function.
Candidate earnings are based on deterministic expected payoffs rather than a random realization of the election outcome. This payment rule removes additional payoff risk while preserving the strategic incentives generated by the winning-probability function. After each election, participants observe their opponent’s platform, both candidates’ winning probabilities, and their own round earnings.
The six treatment cells vary electoral nonresponsiveness and identity-related distortion in the computerized electorate:
1. Rational baseline, lower nonresponsiveness
The computerized electorate evaluates platforms without identity distortion, with α=0.60. This is the baseline condition with a relatively responsive electorate.
2. Rational baseline, higher nonresponsiveness
The electorate remains free of identity distortion, but nonresponsiveness increases to α=0.80. Comparing this cell with Cell 1 identifies how weaker electoral responsiveness affects candidate behavior.
3. Spending–tax frame, lower nonresponsiveness
Identity-related distortions are activated, and fiscal choices are evaluated through the spending–tax frame. Electoral nonresponsiveness is α=0.60.
4. Institutional-accountability frame, lower nonresponsiveness
The same identity distortions and lower nonresponsiveness are used, but platforms are evaluated through the institutional-accountability frame. Comparing Cells 3 and 4 identifies the framing effect when the electorate is relatively responsive.
5. Spending–tax frame, higher nonresponsiveness
Identity distortions operate through the spending–tax frame, with higher nonresponsiveness, α=0.80.
6. Institutional-accountability frame, higher nonresponsiveness
Identity distortions operate through the institutional-accountability frame, also with α=0.80. Comparing Cells 5 and 6 identifies the framing effect when electoral accountability is weaker.
In the two rational-baseline cells, the identity-distortion parameters are set to zero. These cells identify the effect of reduced electoral responsiveness on candidates’ platform choices. In the remaining four cells, the same public-good and rent distortions are activated, and the design crosses the two levels of nonresponsiveness with two alternative electoral frames. This permits comparisons between the spending–tax and institutional-accountability frames at a common level of electoral responsiveness and tests whether frame effects are amplified when the electorate is less responsive.
Before the paid rounds, participants provide informed consent, read the experimental instructions, complete non-incentivized practice tasks, and use the payoff calculator with hypothetical platform pairs. They must pass a comprehension test before proceeding. Incorrect responses generate item-specific feedback and an opportunity to review the relevant instructions.
After the final round, participants complete a post-experiment questionnaire and receive a fixed participation payment plus the incentive payment accumulated during the experiment. The study is implemented in oTree. The software records platform choices, opponent choices, winning probabilities, earnings, calculator queries, decision times, comprehension attempts, timeout events, reconnections, and other implementation indicators.
Hypotheses:
Hypothesis 1: Electoral nonresponsiveness and rent extraction
Candidates are expected to choose higher private rents when the electorate is less responsive to differences between platforms. In the rational-baseline conditions, this means that rents should be higher in Cell 2, where nonresponsiveness is high, than in Cell 1, where nonresponsiveness is low.
When identity distortion is activated, its effect depends on the frame. Under the spending–tax frame, identity distortion should encourage candidates to choose higher rents. Under the institutional-accountability frame, identity distortion should discipline candidates and lead them to choose lower rents. Both effects should be stronger when electoral nonresponsiveness is high. Accordingly, the difference between Cells 5 and 2 should be larger than the difference between Cells 3 and 1, while the rent reduction in Cell 6 relative to Cell 2 should be larger than the rent reduction in Cell 4 relative to Cell 1.
Hypothesis 2: Differences between the two frames
Holding electoral nonresponsiveness constant, candidates should choose higher rents under the spending–tax frame than under the institutional-accountability frame. Therefore, rents should be higher in Cell 3 than in Cell 4 and higher in Cell 5 than in Cell 6.
Because the public-good distortion is held constant across the two frames, differences in total taxation are expected to be driven by differences in rents. Total taxes should therefore also be higher under the spending–tax frame than under the institutional-accountability frame at both levels of nonresponsiveness.
Hypothesis 3: Welfare effects of the frames
Holding electoral nonresponsiveness constant, the institutional-accountability frame should generate at least as much rational welfare as the spending–tax frame. Welfare should therefore be weakly higher in Cell 4 than in Cell 3 and weakly higher in Cell 6 than in Cell 5.
The welfare difference should be strictly positive when the institutional-accountability frame successfully reduces rent extraction. The expected welfare advantage of the institutional-accountability frame should also be larger when electoral nonresponsiveness is high, because the predicted difference in rents between the two frames is stronger in that condition.