Do parties affect public policies?
My new paper with Kilian Seng in @EJPRjournal shows that most existing studies cannot be trusted. Here’s why.
(Thread)
My new paper with Kilian Seng in @EJPRjournal shows that most existing studies cannot be trusted. Here’s why.
(Thread)
Many studies have analyzed whether parties shape policy. In quantitative macro-comparative research, the standard is to use time-series cross-section models to annual observation data. There are tons of these studies.
(Below is an unreadable lit-review in one big table.) (2/n)
(Below is an unreadable lit-review in one big table.) (2/n)
But governments hardly change annually. Usually they change at elections or when new coalitions form. Thus, the number of observations is artificially inflated, leading to incorrect estimates. (3/n)
Some recent work (including Kilian’s and my own in @JEPP) has recognized this problem and suggested using cabinet-terms as the unit of analysis instead.
This makes sense 👏 , but (1) introduces comparability problems as (2) throws away information. (4/n)
This makes sense 👏 , but (1) introduces comparability problems as (2) throws away information. (4/n)
We propose multilevel models as a superior solution. We design a model with annual observations („country-years“) nested in governments, but also nested in countries and time. The model has a couple of advantages spelled out in the paper.
(5/n)
(5/n)
Empirically, we study welfare spending in 23 countries, 6 decades. Welfare is the largest expenditure category in most countries. We study total spending as well as specific sectors (pensions, health care, active labor market policies, unemployment benefits, & education.) (6/n)
What do we find? Maybe most importantly, we show that standard approach used in the existing literature produces overconfident results. While we continue to find partisan effects (mainly stemming from the mid-century) the estimates come with larger uncertainty. (7/n)
Moreover, we detect an interesting conditional effect: It matters how parties talk about welfare in their manifestos - but only for center & right-wing parties. This implies: welfare scholars should pay particular attention to how the political right frames welfare issues.(8/n)
Bottom line: Whenever you are interested in partisan effects on policies or other outputs/outcomes, think carefully about the time dimension.
Multilevel models could easily be applied to any kinds of policy outputs/outcomes and produce more accurate estimates. (9/9)
Multilevel models could easily be applied to any kinds of policy outputs/outcomes and produce more accurate estimates. (9/9)
