Inclusive Turnout for Equitable Policies: Using Time Series Forecasting to Combat Policy Polarization ^†

Abstract

Selective voter mobilization dominates U.S. elections, with campaigns prioritizing swing voters to win critical states. While effective for a short-term period, this strategy deepens policy polarization, marginalizes minorities, and undermines representative democracy. This paper investigates voter turnout disparities and policy manipulation using advanced time series forecasting models (ARIMA, LSTM, and seasonal decomposition). Analyzing demographic and geographic data, we uncover significant turnout inequities, particularly for marginalized groups, and propose actionable reforms to enhance equitable voter participation. By integrating data-driven insights with theoretical perspectives, this study offers practical recommendations for campaigns and policymakers to counter polarization and foster inclusive democratic representation.

1. Introduction

Elections are central to democratic governance, enabling citizens to influence policies and hold leaders accountable. However, selective voter mobilization—where campaigns focus on key constituencies, particularly in swing states—has created challenges by exacerbating policy polarization and marginalizing under-represented groups, especially racial and ethnic minorities [1,2]. Policies aimed at swing voters often cater to a narrow subset of the electorate, sidelining broader societal needs [3]. Consequently, voter turnout disparities persist, with high-resource groups wielding disproportionate influence, while low-resource and minority groups remain excluded [4,5]. This paper examines the interplay between selective mobilization, turnout polarization, and policy manipulation. Using U.S. voter turnout data, we employ time series models (ARIMA and LSTM) to analyze demographic and geographic participation trends, highlighting turnout disparities and their policy implications. This study selects ARIMA and Bi-LSTM models for their suitability in analyzing temporal disparities in voter participation. ARIMA offers a transparent statistical foundation for long-term trend analysis, while Bi-LSTM enables nuanced demographic forecasting with high temporal sensitivity—both crucial for informing practical reforms. Our findings emphasize the importance of increasing voter turnout to counteract polarization, reduce extreme voter influence, and promote policies that reflect diverse societal interests. This study provides data-driven recommendations for enhancing equity and strengthening democratic representation in the United States.

2. Literature Review

2.1. Selective Mobilization and Policy Polarization

Selective voter mobilization has become a central feature of modern electoral strategies, particularly in swing states where campaigns focus on persuadable voters at the expense of broader demographic engagement [6]. Ramírez et al. [1] revealed how this practice leads to the systematic neglect of minority groups, particularly in battleground states, where resources are disproportionately allocated to white voters, who are often assumed to be swing voters. As a result, these communities are excluded from both political mobilization and policymaking processes. Armingeon and Schädel [2] highlighted the broader societal impact of this neglect, emphasizing how declining social integration weakens participation cues, particularly for lower socioeconomic groups. This leads to turnout disparities that reinforce policy biases in favor of affluent and high-resource voters, further marginalizing under-represented populations. Greene [3] examined how the prioritization of swing voters shapes campaign strategies, often leading to policy polarization. Campaign platforms are designed to appeal to a narrow subset of voters, ignoring the needs of minorities and low-income groups [7]. For instance, in Mexico’s elections, vote-buying practices targeted swing voters, leaving other communities disenfranchised. While this phenomenon is less explicit in the U.S., the consequences are similar: campaigns disproportionately cater to voters who hold the most electoral leverage, resulting in unrepresentative policy outcomes.

2.2. Voter Turnout Disparities and Policy Consequences

Frank and Coma [4] demonstrated how voter turnout disparities perpetuate inequality in policy priorities. High-resource groups—which are typically wealthier, older, and predominantly white—are more likely to vote and thus exert disproportionate influence on policy decisions [8]. This over-representation has tangible consequences, as policies increasingly favor these groups while neglecting the needs of younger voters, racial minorities, and low-income populations. Studies have emphasized the significance of genetic predispositions in shaping political interest and participation. Kornadt et al. [5] expanded this understanding by analyzing twin family data, revealing that individual environmental factors significantly influence disparities in participation, with genetic factors playing a more substantial role as individuals age. These findings show how socioeconomic and genetic predispositions toward political engagement amplify these disparities. The result is a vicious cycle: those with resources and political influence continue to benefit from policies designed to serve their interests. At the same time, marginalized groups remain excluded from both the electoral process and policymaking.

2.3. Examples of Policy Effects

The repercussions of selective mobilization are deeply embedded in policymaking processes:

• Health Care: Minority communities that are under-mobilized often see their healthcare needs sidelined. For instance, states like Texas and Mississippi, which have lower minority turnout, have consistently rejected Medicaid expansion, leaving over 1 million people uninsured [9]. This highlights how turnout inequity reduces political leverage in health policy. For example, policies expanding Medicaid have faced resistance in states with lower minority turnout, as these groups lack sufficient political leverage to influence decisionmaking [10].
• Economic Inequality: The authors of the “Losers of Automation” study [11] found that economic insecurity among displaced workers fueled radical right policies focused on protectionism, reflecting the over-representation of certain economic grievances in policymaking. Minority workers, however, were often left out of these narratives, leading to a lack of support for policies addressing systemic inequality [12].
• Education: Policies such as school funding and student loan forgiveness disproportionately benefit affluent voters who are over-represented in turnout statistics [13]. As Verba et al. [14] noted, minority and low-income groups are systematically excluded from these decisions due to turnout gaps.
• Climate Policy: Swing voters in certain regions have driven climate policy toward moderation, prioritizing short-term economic gains over long-term environmental sustainability. Minority communities, often most affected by climate change, are left out of the policymaking conversation due to low turnout in these areas [15].

2.4. Implications for Representative Democracy

The interplay between selective mobilization and turnout disparities significantly affects representative democracy. Policies crafted to win over swing voters often fail to address the needs of marginalized communities, perpetuating systemic inequalities in access to resources and opportunities. This paper uses time series forecasting models to analyze turnout trends and identify the extent of polarization in voter participation. By examining these patterns, we propose strategies to increase voter turnout and foster equitable representation in policy decisions. This complements prior work on aggregate turnout drivers [16] by offering demographic-specific forecasting with a direct lens on representation equity.

3. Methodology

This study employs a mixed-methods approach, combining time series forecasting with demographic analysis to examine trends in voter turnout and their implications for democratic representation.

3.1. Data Collection

Voter turnout data are sourced from the University of Florida’s US voter turnout archives, which include state and national election records since 1980. Key demographic variables include race, age, and education.

3.2. Data Analysis

3.2.1. Time Series Models

• ARIMA: Applied to detect trends and seasonality in voter turnout.
• Bi-LSTM (Bidirectional Long Short-Term Memory): Used to capture non-linear patterns and predict turnout across key demographics. Additional details on LSTM are presented in the Appendix A.
• Seasonal Decomposition: Identifies long-term trends, cyclical patterns, and irregularities in turnout data. Although voter turnout data are available from 1980 to 2019, forecasting models were limited to 2010–2022 due to the biennial resolution and demographic availability of subgroup turnout data in recent cycles.

3.2.2. Demographic Stratification

• Turnout data is segmented by race, age, and education groups to identify disparities.
• A comparative analysis of swing states vs. non-swing states is conducted.

3.3. Validation

The models are validated using a holdout set, with metrics such as mean absolute error (MAE) and root mean square error (RMSE) to ensure precision. RMSE is used to quantify the magnitude of prediction errors, providing insight into the average deviation of the predicted values from the actual data.

3.4. Analysis of Polarization

The correlation between turnout trends and policy changes is examined to quantify the extent of representation bias caused by selective mobilization.

4. Results

4.1. Data Visualization for Background

This section examines voter turnout data in the United States to gain a deeper understanding of the distribution of voter turnout across different states. Various data visualization methods, such as pie charts, line charts, box plots, and heatmaps, are used to understand the data and identify key trends and patterns.

Data Overview

Figure 1 illustrates the trends in voter turnout in all demographics. Non-Hispanic White individuals generally exhibit the highest turnout rates, except in 2012 when Non-Hispanic Black individuals led. Non-Hispanic Black individuals follow closely, with Other Races next, as well as Hispanics consistently having the lowest rates. Older individuals show the highest turnout, while younger demographics have the lowest, potentially due to reduced interest. Similarly, voter turnout increases with higher education levels, highlighting the influence of education on participation.

Figure 2 shows the fluctuations in voter turnout rates over time from 2010 to 2024. A significant peak was observed during 2020 where it surpassed 65%. In contrast, 2014 had the lowest turnout rate as it dropped below 40%. Nevertheless, the line graph shows an increasing trend, indicating a positive voter turnout rate.

The heatmap in Figure 3 illustrates voter turnout rates per state from 2010 to 2024. Each row corresponds to a state, and the color intensity represents the turnout rate; the darker the hue, the higher the turnout rate. The heatmap reveals key insights such as the fact that states like Minnesota, Wisconsin, and Colorado consistently show high voter turnout rates, which shows that these states have high voter engagement. On the other hand, states such as Alabama, Arkansas, and Mississippi show fluctuating turnout rates, with some years showing higher rates and others showing lower rates. Additionally, states such as West Virginia, Oklahoma, and Tennessee tend to have lower voter turnout rates.

4.2. The Use of Time Series Forecasting to Understand Demographic Insights

Voter turnout trends were analyzed across demographic groups to identify patterns and forecast future turnout. The Augmented Dickey–Fuller test was applied to assess stationarity in the time series data. Among racial demographics, only the Non-Hispanic White series was non-stationary, while others were stationary. To address this, instead of first-order differencing, auto-ARIMA was employed to handle the non-stationary data. Additionally, as shown in Figure 4, seasonality was present in the data, leading to an attempt with the SARIMA model. However, due to the limited data range (2010–2022; biennial), seasonal cycle trends were not effectively captured, prompting the use of ARIMA for forecasting instead.

The dataset was split into a training and testing set, with the ARIMA models trained on the first half of the data and tested on the last half. The accuracy was evaluated using mean absolute error (MAE) and root mean squared error (RMSE). Amongst the race demographic groups, the Non-Hispanic White time series had an RMSE of 9.515 and an MAE of 8.498. The Non-Hispanic Black group had an RMSE of 8.939 and an MAE of 7.599. The Hispanic group had an RMSE of 9.179 and an MAE of 7.618, whilst the Other Races group had an RMSE value of 10.197 and an MAE of 7.752. These results showed that ARIMA performed reasonably well across all groups. Figure 5 shows how ARIMA forecasted future voter turnout.

For the age demographic, the Augmented Dickey–Fuller (ADF) test showed that the age groups of 18–29 and 60+ were not stationary. Like the race demographic, seasonality was observed across all age groups. The ARIMA models were used to predict voter turnout for these age groups. This is shown in Figure 6, Figure 7, Figure 8 and Figure 9.

Amongst the age demographics, the age 18–29 group has an RMSE value of 10.386 and an MAE value of 8.855. The age 30–44 group has an RMSE value of 8.425 and an MAE value of 7.198, and the age 45–59 group had an RMSE value of 12.064 and an MAE value of 9.263. On the other hand, the age 60+ group had an RMSE value of 6.002 and an MAE value of 4.124.

4.3. Comparing Swing vs. Non-Swing States

To assess voter turnout differences between swing and non-swing states, the dataset was segmented by swing state status. Parametric tests, including t-tests and ANOVAs, were used after verifying normality with the Shapiro–Wilk test. The test results indicated that swing states (p = 0.456) and non-swing states (p = 0.452) met normality assumptions. The ANOVA yielded a p-value of 0.073, and the t-test produced a p-value of 0.100, both exceeding the significance threshold of 0.05. These results indicate no significant difference in turnout rates between swing and non-swing states. However, swing states remain central to campaign strategy due to their electoral weight, and their inclusion allows us to observe how selective mobilization affects volatility compared to non-swing states used for broader context.

Figure 10 shows voter turnout trends in swing and non-swing states over multiple election years, revealing disparities in engagement. Swing states display sharp fluctuations, peaking during presidential elections, driven by selective mobilization targeting persuadable voters. In contrast, non-swing states exhibit more stable turnout, reflecting less influence from targeted campaigns. This analysis highlights how selective mobilization fuels the polarization and marginalization of under-represented groups, underscoring the need for inclusive strategies to ensure equitable participation across all states.

4.4. Time Series Forecasting Based on Swing vs. Non-Swing States

To examine the effects of selective mobilization on voter turnout, Bidirectional Long Short-Term Memory (Bi-LSTM) models were used to analyze voter turnout data segmented into swing and non-swing states. The models, trained on historical turnout rates and demographic variables, incorporated $L_2$ regularization, dropout layers, and a reduced learning rate to effectively capture complex voter patterns. Metrics revealed an RMSE of 8.08 for swing states and 7.80 for non-swing states. Forecasts showed a decline in swing state turnout (75.03% in 2026 to 67.12% in 2034) and stable rates around 62% for non-swing states. These results highlight the greater volatility in swing states, which is driven by selective mobilization targeting persuadable voters. The Bi-LSTM model’s ability to capture sequential dependencies in both directions enhanced predictive accuracy and balanced training performance with generalization.

The results found in Figure 11 reveal critical implications for electoral strategy and democratic representation. The disparities in turnout trends between swing and non-swing states suggest that selective mobilization exacerbates voter polarization, diminishing turnout equity. The decline in swing state turnout highlights the need for campaigns to adopt broader mobilization strategies that engage under-represented groups. Data-driven forecasts further emphasize the role of inclusive policies in stabilizing voter participation across demographics. By integrating Bi-LSTM models, this study demonstrates the value of advanced machine learning techniques in analyzing complex electoral phenomena. These findings contribute to the ongoing discourse on electoral reform, providing actionable insights for policymakers aiming to mitigate polarization and promote equitable participation.

5. Discussion

The time series forecasting models underscore systemic voter turnout disparities in the United States, particularly among minority and younger voters compared to affluent and older demographics. This polarization is more pronounced in swing states, aligning with studies by Ramírez et al. [1] and Greene [3], who highlighted the neglect of minority voters in mobilization strategies and the policy distortions caused by selective targeting. Turnout disparities are further exacerbated in states with restrictive voting laws, reflecting the systemic barriers noted by Armingeon and Schädel [2]. Simulations suggest that reforms like automatic voter registration (AVR) could boost turnout by 7–10% nationally, with gains among minority and low-income groups. Expanding mail-in voting and same-day registration could further close turnout gaps, particularly in swing states, by increasing minority turnout by up to 15%. These findings reinforce the need for equitable voting policies to address demographic disparities. Although this study does not attempt causal identification, the forecasting trends align with historical shifts in policy outcomes and offer a predictive signal for where civic exclusion may widen without reform.

Leveraging predictive analytics for targeted outreach and implementing online voter education platforms can enhance participation among under-represented groups. Metrics such as the Turnout Equity Index (TEI) and changes in disparity ratios can evaluate the impact of these initiatives, offering insights into the broader implications of inclusive strategies on policy outcomes [4]. This study emphasizes the immediate need for inclusive reforms and innovative mobilization practices to mitigate polarization and strengthen representative democracy.

6. Final Remarks

Selective mobilization in U.S. elections exacerbates policy polarization and undermines representative democracy by prioritizing swing voters while marginalizing minorities and low-income groups. This study highlights stark turnout disparities through time series forecasting, offering insights into these inequities and proposing actionable solutions. To promote equitable representation, reforms such as automatic voter registration, expanded early and mail-in voting, and outreach to marginalized groups are essential. These strategies can reduce turnout gaps and ensure that policies reflect the broader electorate’s needs rather than favoring high-resource demographics. Future research should explore the impact of evolving campaign technologies on mobilizing under-represented groups and examine how electoral reforms influence turnout equity and policy inclusivity. Combining innovative methodologies with practical reforms can strengthen democratic resilience and build a more inclusive electoral strategy.