Difference between revisions of "Quasi-Experimental Methods"
| (One intermediate revision by the same user not shown) | |||
| Line 1: | Line 1: | ||
Quasi-experimental methods are research designs that that aim to identify the impact of a particular intervention, program or event (a "treatment") by comparing treated units (households, groups, villages, schools, firms, etc.) to control units. While quasi-experimental methods use a control group, they differ from [[Experimental Methods | experimental methods]] in that they do not use [[Randomization | Quasi-experimental methods are research designs that that aim to identify the impact of a particular intervention, program, or event (a "treatment") by comparing treated units (households, groups, villages, schools, firms, etc.) to control units. While quasi-experimental methods use a control group, they differ from [[Experimental Methods | experimental methods]] in that they do not use [[Randomization | randomization]] to select the control group. Quasi-experimental methods are useful for estimating the impact of a program or event for which it is not [[Research Ethics | ethically]] or logistically feasible to '''randomize'''. This page outlines common types of quasi-experimental methods. | ||
==Read First== | ==Read First== | ||
*Common examples of quasi-experimental methods include [[Difference-in-Differences | | *Common examples of quasi-experimental methods include | ||
*In general, quasi-experimental methods require larger sample sizes and more assumptions than experimental methods in order to provide valid and unbiased estimates of program impacts. | **[[Difference-in-Differences | Difference-in-differences]] | ||
**[[Regression Discontinuity | Regression discontinuity design]] | |||
**[[Instrumental Variables | Instrumental variables]] | |||
**[[Propensity Score Matching|Propensity score matching]] | |||
**[[Matching | Matching]] | |||
*In general, quasi-experimental methods require larger [[Sampling|sample sizes]] and more assumptions than [[Experimental Methods|experimental methods]] in order to provide valid and unbiased estimates of program impacts. | |||
== Overview == | == Overview == | ||
Like experimental methods, quasi-experimental methods aim to estimate program effects free of confoundedness, reverse causality or simultaneous causality. While quasi-experimental methods use a counterfactual, they differ from | Like [[Experimental Methods|experimental methods]], quasi-experimental methods aim to estimate program effects free of confoundedness, reverse causality or simultaneous causality. While quasi-experimental methods use a counterfactual, they differ from '''experimental methods''' in that they do not [[Randomization | randomize]] treatment assignment. Instead, quasi-experimental methods exploit existing circumstances in which treatment assignment has a sufficient element of randomness, as in [[Regression Discontinuity | regression discontinuity design]] or [[Event Study|event studies]]; or simulate an experimental counterfactual by constructing a control group as similar as possible to the treatment group, as in [[Propensity Score Matching|propensity score matching]]. | ||
==Assumptions and Limitations== | ==Assumptions and Limitations== | ||
In general, quasi-experimental methods require larger samples than experimental methods. Further, for quasi-experimental methods to provide valid and unbiased estimates of program impacts, researchers must make more assumptions about the control group than in experimental methods. For example, difference-in-differences relies on the equal trends assumption | In general, quasi-experimental methods require larger [[Sampling|samples]] than [[Experimental Methods|experimental methods]]. Further, for quasi-experimental methods to provide valid and unbiased estimates of program impacts, researchers must make more assumptions about the control group than in '''experimental methods'''. For example, [[Difference-in-differences|difference-in-differences]] relies on the equal trends assumption, while [[Matching|matching]] assumes identical unobserved characteristics between the treatment and control groups. | ||
== Additional Resources == | == Additional Resources == | ||
Latest revision as of 19:07, 9 August 2023
Quasi-experimental methods are research designs that that aim to identify the impact of a particular intervention, program, or event (a "treatment") by comparing treated units (households, groups, villages, schools, firms, etc.) to control units. While quasi-experimental methods use a control group, they differ from experimental methods in that they do not use randomization to select the control group. Quasi-experimental methods are useful for estimating the impact of a program or event for which it is not ethically or logistically feasible to randomize. This page outlines common types of quasi-experimental methods.
Read First
- Common examples of quasi-experimental methods include
- In general, quasi-experimental methods require larger sample sizes and more assumptions than experimental methods in order to provide valid and unbiased estimates of program impacts.
Overview
Like experimental methods, quasi-experimental methods aim to estimate program effects free of confoundedness, reverse causality or simultaneous causality. While quasi-experimental methods use a counterfactual, they differ from experimental methods in that they do not randomize treatment assignment. Instead, quasi-experimental methods exploit existing circumstances in which treatment assignment has a sufficient element of randomness, as in regression discontinuity design or event studies; or simulate an experimental counterfactual by constructing a control group as similar as possible to the treatment group, as in propensity score matching.
Assumptions and Limitations
In general, quasi-experimental methods require larger samples than experimental methods. Further, for quasi-experimental methods to provide valid and unbiased estimates of program impacts, researchers must make more assumptions about the control group than in experimental methods. For example, difference-in-differences relies on the equal trends assumption, while matching assumes identical unobserved characteristics between the treatment and control groups.
Additional Resources
- Robert Michael's slides on Quasi-Experimental Designs
- Gertler et al.’s Impact Evaluation in Practice