Economic Development, Inequality and Dynamics of Social Movements in the United States: Theory and Quantitative Analysis

Abstract

How have social movements in the United States been impacted by simultaneously evolving economic realities such as episodes of development and inequality across time? This paper empirically examines how the structural forces of the economy such as growth (income per-capita) and decline (income inequality) interact with the regional characteristics to derive patterns of social movements in United States from 1960 to 1995. I suggest that—unlike the arguments found in popular social movement theories such as relative deprivation and economic grievances that the society will express resentment against lack of financial resources through protesting and riots—there will be less collective action formations during heightened inequality even when there is growth in per-capita income. This paper provides novel application of methodological approaches in social movement studies such as the Generalized Additive Models with smoothing functions and Synthetic Control Method to extract micro-level inferences on the relationship between economic factors and social movement formations. I gauge the implications of the main argument with a new dataset that is a composition of aggregated levels of social movements per-capita, real per-capita personal income, income inequality index, labor unemployment laws, social policy liberalization index and equal pay laws among other variables. The empirical exercises reveal that when accounting for the full range of socio-economic variables with fixed effects and instrumental variables, the dual impact of economic growth and decline on social movements is non-linear and U-shaped in the US states across time.

Appendices

Appendix 1: Data

See Figs. 14, 15 and 16.

Fig. 14

COSP 1960–1965, averages of various economic variables at neighboring States. Notes: Calculations are author’s own. COSP correlates of state policy dataset. All variables are in logarithmic form

Fig. 15

Density plots for NSPC summarized by various policies at the state level. Notes: Calculations are author’s own. NSMPC normalized social movements per-capita. Yes and No indicate whether or not a policy is at place in each state-year dyad

Fig. 16

Correlation matrix between instrumented variables used in the second stage. Notes: V1 = Norm TCAPC, V2 = TCAPC, V3 = RPCPI (I), V4 = Income Ineq. (I), V5 = General Rev. (I), V6 = Labour Unemp. Comp. (I), V7 = Violent crime rate, V8 = Social liberalism, V9 = Social Policy Liber. The term (I) = instrument in the 1st stage

Appendix 2: Methodology

The SCM has various advantages over other empirical models used with cross-country time series longitudinal datasets. First, it allows to predict not only immediate influence of independent variable on levels in the outcome variable but also aggregate output impact for every single year for each country that is available. Second, the SCM considers the untreated observations in cross-sectional time series format by applying weighted means of other units as the counterfactuals (Gobillon and Magnac 2016). The SCM achieves this by composing control groups consisting of weighted combination of observation units (in our case countries-years) and comparing them to treatment groups which have some characteristic that is absent in the control groups (Xu 2017). Third, in contrast to DiD method and fixed or random effects techniques popular in panel data analysis that only control for time-invariant influences, the SCM allows to account for time-variant unobservable factors (Abadie and Gardeazabal 2003). Fourth, compared to system generalized method of moments and other dynamic models that permit to estimate the average effect of treatment drawing on the entire sample (Arellano and Bond 1991), the SCM enables to analyze the causal effect of regime instabilities (such as corruption and income declines) on social movement levels in temporal perspective for treatment groups within individual units (states) of our data.

In this manner, I leverage the comparative case study capability of SCM approach to enable fabricating a synthetic control group to utilize by paralleling the outcome of interest, NSMPC , in a treated unit with the results of that control group without the explicit disclosure of that control group to the intervention of interest (Gilchrist et al. 2022). Given the intuition and setup explain for Eq. (9) in “Case Studies Using Synthetic Control Method”, let’s represent the matrices of treatments and outcomes produced by those treatments with a combination of $\textit{N} \times \textit{T}$ matrices and denote this combination by M. Hence, referring to the Eq. (9), I suggest that the outcome matrix of interest is denoted by Y given the structure of M which arrives to:

$$\begin{aligned} Y_{i,t} = M_{i,t} Y_{i,t} (1) + (1- M_{i,t}) Y_{i,t}(0). \end{aligned}$$

(10)

From the equation above, I consider that the influence of treatment is expressed for each period t, suggesting that it is subject to change temporally. The primary empirical task becomes the requirement to estimate $Y_{j,t N}$ while accounting for the scenario of where the outcome is not treated. To achieve this objective, I recognize that the cluster of units in the donor pool could aggregate the features of the treated unit more precisely than the potential effect produced by units that are not treated. This property of the SCM model given the input from the donor pool establishes the method as a weighted average of the units pooled from the donor group comprising the collection of controls. Given the combination of time and units matrices M from equation 10, I have weights defined as M = $(m2,\ldots ,mjt+1)$ which encompasses the synthetic control estimate of $Y_{j,t}^{N}$ in the following equation:

$$\begin{aligned} \hat{Y}_{j,t N} = \sum _{j=2}^{J+1} m_{j}Y_{j,t}. \end{aligned}$$

(11)

Appendix 3: Empirical Outcomes

See Figs. 17, 18, 19 and 20 and Tables 8, 9, 10, 11 and 12.

Fig. 17

Second-stage Poisson GAM baseline model validity checks. Notes: Author’s own calculation. The RPCPI and income inequality variables’ values are predicted in the first stage

Fig. 18

Second-stage Poisson GAM state FE model validity checks. Notes: Author’s own calculation. The RPCPI and income inequality variables’ values are predicted in the first stage

Fig. 19

Second-stage Poisson GAM year FE model validity checks. Notes: Author’s own calculation. The RPCPI and income inequality variables’ values are predicted in the first stage

Fig. 20

Second-stage Poisson GAM year FE model with additional controls validity checks. Notes: Author’s own calculation

Table 8 Results of the second stage Poisson GAM baseline model

Table 9 Results of the second stage Poisson GAM state FE model

Table 10 Results of the second stage Poisson GAM year FE model

Table 11 Results of the second stage Poisson GAM year FE model

Table 12 Contribution of each donor group unit to the synthetic data for treatment