Democratic Trajectories in the Third Wave: Aligning Theory and Methods
Democratic Trajectories in the Third Wave: Aligning Theory and Methods (DOI: https://doi.org/10.1017/S0007123425000237) was written by Aníbal Pérez-Liñán and Scott Mainwaring in 2025. It was published in the British Journal of Political Science (vol. 55).
The authors essentially make the case for using latent growth curves to model democratization, to better align with qualitative research/philosophy.
A prevalent quantitative method for studying democratic trajectories is to regress structural predictors on country-year measurements of democratization. Most studies focus on economic growth as the predictor. The dependent variable is generally selected from Polity, Freedom House, or V-Dem. There are two schools of thought for the data generating process that should be modeled:
Level of democracy is 'sticky'. A lagged dependent variable term is introduced to reflect lasting effects. dt = gt + dt-1; dt = d0 + g̅*t.
Conditioning of growth by democracy. Growth is itself modeled in terms of lagged democracy. dt = d0 + gt; gt = dt-1 + ut; dt = d0(1+t) + u̅*t.
In either school, implicit assumption is that democracy is stable ("stationary") absent exogenous forces.
Qualitative studies variably look at countries or regimes. Either way, these studies often use comparative politics methods and focus on critical junctures. Generally speaking, these studies look at much longer term causation than the quantitative ones, and assume that there is an inherited ("path dependent") state of the world dictating the direction of democratization ("reversionary level of democracy").
The authors argue that the prevalent quantitative methods are fundamentally flawed for failing to match qualitative studies in these important ways.
From there, the authors walk through the intuition of fixed effect models, then random effect/hybrid models, and finally demonstrate the equivalence of random effects and latent variables.
This leads into the introduction of latent growth curves. These feature a latent variable for a trajectory (i.e., T) that is designed to carry a linearly-increasing loading across measurements (i.e., [1, 2, 3...]). The authors also demonstrate using quadratic and cubic terms (i.e., T2 and T3) for nonlinear trajectories.
The authors then model on spells of democracy (i.e., annual measurements from establishment to disestablishment, normalized to start at period 1) from 1974 to 2017. Authors use V-Dem to classify democracy from autocracy. Predictors of within effects include some of the components of V-Dem itself (i.e., Gini coefficient) and GDP transformations (i.e., year over year growth rate, per capita level). Between effects and latent variables are predicted by means (i.e, average GDP growth rate) Of the 103 spells, 98 have complete information and are included in analysis.
The authors primarily evaluate the models using visualizations (i.e., graphing the latent growth curve's prediction against those of the fixed effects and hybrid models) and RMSE. The exact model is largely not statistically significant and is not a serious candidate for use.
Reading notes
It's an interesting cross-disciplinary article.
I am interested in reading more on panel analysis using normalized spells vs. actual timed measurements. I suppose the implicit assumption is that age of a democracy has more exogeneity than year. Is that fair? I don't know.
I also find it's unfortunate to discard all information about the 'trajectory' of these predictors during autocratic regimes. That would be an important component, I have to imagine.