Quasi-Experimental Research

Quasi-experimental research is used to investigate causal relationships when researchers cannot randomly assign individuals, organisations, locations or other units to treatment and control conditions. Like randomized experiments, quasi-experiments are concerned with estimating the effects of interventions, treatments, policies or exposures. Their defining challenge is that treatment assignment occurs through a non-random process, which makes it more difficult to determine whether observed differences were actually caused by the treatment.

The absence of randomization does not mean that causal inference is impossible. Instead, quasi-experimental research relies on alternative strategies for constructing a credible comparison between what happened after an intervention and what would probably have happened without it. The strength of the resulting causal claim depends on the particular research design, the assumptions required by that design, and the extent to which plausible alternative explanations have been addressed.

This makes quasi-experimental research especially important in business, economics, public policy and other applied fields where interventions frequently occur in real-world settings but researchers cannot control who receives them. A company may introduce a new technology in selected branches, a government may implement a policy in one region before another, or eligibility for a programme may depend on a predetermined threshold. Such situations can provide opportunities for causal research when the assignment process and available comparison are methodologically appropriate.

On this page:

• Quasi-Experimental Research Explained Simply
• What Is Quasi-Experimental Research?
• The Counterfactual Problem
• True Experiments vs Quasi-Experiments vs Observational Studies
• Major Quasi-Experimental Designs and Methods
• How to Design Quasi-Experimental Research
• Dudovskiy Quasi-Experimental Causal Claim Framework
• Threats to Causal Inference
• Application of Quasi-Experimental Research: an Example
• Advantages and Limitations of Quasi-Experimental Research
• Common Mistakes in Quasi-Experimental Research
• Quasi-Experimental Research in Business Research
• Quasi-Experimental Research in the Age of AI and Digital Research
• When to Use Quasi-Experimental Research
• Dissertation Example
• Exam Tip

Methodological feature	Randomized experiment	Quasi-experiment	Observational study
Main purpose	Estimate causal effects	Estimate causal effects	Describe relationships; may also investigate causal questions
Random assignment	Yes	No	No
Treatment or intervention	Deliberately assigned	May be introduced or arise through policies, thresholds, programmes or other processes	Usually naturally observed
Counterfactual strategy	Randomized control group	Design-specific comparison or source of variation	Design and statistical adjustment
Central causal challenge	Implementation, attrition, non-compliance and other validity threats	Non-random assignment and design-specific assumptions	Confounding and selection
Potential for causal inference	Strong when well implemented	Can be strong when identifying assumptions are credible	Depends heavily on design and assumptions

These categories are useful for understanding different research-design logics, but their boundaries should not be interpreted as a simple hierarchy of research quality. The credibility of a causal conclusion ultimately depends on the particular design, its implementation and the assumptions connecting the observed evidence to the causal claim.

Quasi-Experimental Research Explained Simply

Imagine a supermarket chain introduces a new employee training programme in 20 stores while another 20 stores continue operating without it. Six months later, customer satisfaction has increased substantially more in the stores that received the training.

It would be tempting to conclude that the training programme caused the improvement. However, management may have selected those stores precisely because their managers were particularly motivated, because customer satisfaction had previously been unusually low or because the stores were located in regions where other improvements were also being introduced. Any of these differences could contribute to the subsequent results.

A randomized experiment would address much of this problem by randomly assigning stores to the training and control conditions. When randomization is unavailable, quasi-experimental research tries to establish a credible comparison in another way. The researcher might examine differences between the groups before the programme, compare changes over time, exploit a rule that determined which stores received training, or use another design capable of approximating what would have happened to the treated stores without the programme.

The central question is therefore not simply whether customer satisfaction increased after training. It is what would probably have happened to customer satisfaction in those same stores if the training had not been introduced. Quasi-experimental research is fundamentally concerned with making that comparison credible.

Dudovskiy Research Assistant

Not sure if quasi-experimental research is the correct choice for your dissertation?

Enter your topic to receive a free, tailored methodology preview.

Dissertation topic

Generate Free Methodology Preview → Free · No account required

Learn more about Dudovskiy Research Assistant

What Is Quasi-Experimental Research?

Quasi-experimental research investigates causal effects without randomly assigning research units to treatment conditions. This distinguishes it from randomized experiments while retaining the broader experimental objective of determining whether an intervention, treatment or exposure produces a change in an outcome.

Random assignment is valuable because, when properly implemented, it makes treatment assignment independent of pre-existing characteristics on average. Treatment and control groups may still differ by chance, particularly in small samples, but randomization provides a known assignment mechanism that supports causal comparison. Quasi-experimental researchers do not receive this protection automatically because treatment may instead depend on eligibility rules, individual choices, organisational decisions, geographical boundaries, policy changes, timing or other non-random processes.

Consequently, the methodological burden shifts towards understanding the assignment mechanism and identifying another basis for constructing the counterfactual. Different quasi-experimental designs solve this problem differently. Regression Discontinuity Design exploits a treatment threshold, Difference-in-Differences compares changes between groups over time, Interrupted Time Series examines changes in an outcome’s level or trajectory around an intervention, Instrumental Variables use an external source of treatment variation under specific assumptions, and propensity-score methods attempt to improve comparability between groups on observed characteristics.

The term quasi-experimental therefore does not by itself establish that a study provides strong causal evidence. Researchers need to explain what feature of the design makes the treatment and counterfactual comparison informative and what assumptions are required for the resulting estimate to have a causal interpretation.

The Counterfactual Problem

Causal research contains a fundamental difficulty: researchers cannot observe the same unit simultaneously under treatment and non-treatment conditions. Suppose an employee receives a leadership-development programme and subsequently achieves a performance score of 90. To determine the programme’s individual causal effect with certainty, we would need to know what that same employee’s performance would have been at the same time, under otherwise identical circumstances, had the programme not been received. The score might have been 80, suggesting a positive treatment effect, or it might have been 90 anyway, suggesting no effect.

Because both potential outcomes cannot be observed simultaneously for the same individual, researchers need to construct a credible counterfactual using other observations. Modern causal inference formalises this problem through the potential-outcomes framework: each unit has a potential outcome under treatment and another under control, while only the outcome corresponding to the condition actually experienced can be observed.

Randomized experiments address this problem through random assignment, which allows treatment and control groups to provide counterfactual information for one another on average. Quasi-experimental designs lack randomization and therefore rely on alternative sources of comparison or variation. The credibility of the resulting causal inference depends heavily on how convincing that alternative strategy is.

This is also why measuring an outcome before and after an intervention is usually insufficient for establishing causality. A change over time demonstrates that something changed, but without a credible counterfactual it does not establish whether the intervention caused that change.

True Experiments vs Quasi-Experiments vs Observational Studies

A true or randomized experiment deliberately assigns units to treatment conditions using a random mechanism. If the experiment is properly designed and implemented, randomization helps ensure that treatment status is not systematically determined by characteristics that could also influence the outcome. This provides a strong basis for estimating the causal effect of the intervention.

A quasi-experiment lacks random assignment but contains another design feature that can support causal comparison. Treatment might be determined by a cutoff, introduced in one jurisdiction but not another, implemented at a known point in time or affected by an external source of variation. The researcher attempts to exploit this structure to estimate what would have happened without treatment.

An observational study examines exposures, behaviours or characteristics that occur without experimental assignment. Such studies can also investigate causal questions, particularly when supported by careful design and modern causal-inference methods, but confounding and selection require explicit consideration.

These categories should not be converted into a simplistic quality ladder. A poorly implemented randomized experiment can produce misleading evidence, while a carefully designed quasi-experiment can support a persuasive causal argument. The more useful question is what creates the comparison that allows the researcher to estimate the relevant counterfactual and what assumptions make that comparison credible.

Major Quasi-Experimental Designs and Methods

Quasi-experimental research encompasses several approaches that obtain causal leverage in fundamentally different ways. Understanding those differences is more important than memorising a list of techniques because each design addresses the absence of randomization through a particular identification strategy.

Regression Discontinuity Design (RDD) is used when treatment or eligibility changes at a defined threshold on an assignment variable. A scholarship might be awarded to applicants scoring above a specified examination score, for example. Researchers can compare observations immediately around the threshold on the logic that units just above and below it may be highly similar except for their treatment status. The credibility of the design depends on assumptions concerning the assignment process, continuity around the cutoff and the possibility of manipulation of the running variable.

Difference-in-Differences (DiD) compares changes over time between a treatment group and a comparison group. Rather than interpreting the treatment group’s before-and-after change as the intervention effect, DiD examines whether that group changed differently from an untreated comparison group. Its causal interpretation depends critically on whether the comparison provides credible information about how the treated group’s outcome would have evolved without treatment.

Interrupted Time Series (ITS) uses repeated observations before and after an intervention to determine whether its introduction coincides with a change in the level or trajectory of an outcome. A long pre-intervention series can help distinguish an intervention-related change from an underlying trend, while the addition of a suitable comparison series can strengthen the design further.

Instrumental Variables (IV) use variation generated by an instrument to identify a treatment effect under specific assumptions. A valid instrument must provide variation in treatment while satisfying demanding conditions concerning its relationship with the outcome and relevant confounding processes. Instrumental-variable analysis can therefore address certain forms of unobserved confounding, but finding and defending a genuinely credible instrument is often difficult.

Propensity Score Methods, including Propensity Score Matching (PSM), use observed pre-treatment characteristics to improve comparability between treated and untreated groups. PSM can help construct groups with similar distributions of measured covariates, reducing bias associated with those observed characteristics. It cannot automatically eliminate bias caused by relevant characteristics that were not measured, which makes the quality of covariate selection, overlap and balance particularly important.

These methods should not be treated as interchangeable statistical procedures. RDD derives causal leverage from a threshold, DiD from comparative changes, ITS from temporal structure, IV from an instrument and PSM from comparability on observed characteristics. Selecting among them therefore requires understanding how treatment was assigned and what information is available for constructing the counterfactual.

How to Design Quasi-Experimental Research

A strong quasi-experimental study begins with a clearly specified causal question. The researcher should identify the treatment or exposure, the outcome that might be affected, the relevant population and the causal effect of interest before choosing an analytical technique. Starting with available software or a preferred method risks forcing the research question into a design that does not match the actual treatment-assignment process.

The next task is to understand why some units received the treatment while others did not. Treatment might depend on an eligibility threshold, organisational choice, policy jurisdiction, phased implementation, historical circumstances or individual self-selection. Examining this process often reveals both the strongest potential identification strategy and the most important threats to causal inference. A predetermined threshold might suggest RDD, whereas a policy affecting one group but not another at a particular time might create an opportunity for Difference-in-Differences.

Researchers then need to specify the counterfactual strategy explicitly. A comparison group should not be selected simply because its data are readily available; it must provide meaningful evidence about what would probably have happened to treated units without the intervention. The suitability of this comparison is a substantive methodological question, not merely a statistical one.

Once the design has been established, its identifying assumptions should be stated and investigated as far as possible. Diagnostic tests, placebo analyses, falsification exercises, sensitivity analyses and alternative specifications can provide evidence about particular assumptions, although some assumptions cannot be conclusively verified from observed data. The final causal claim should therefore reflect both the strengths of the design and the remaining uncertainty.

Dudovskiy Quasi-Experimental Causal Claim Framework

The Dudovskiy Quasi-Experimental Causal Claim Framework provides a structured way to design and evaluate the reasoning behind a quasi-experimental causal claim. It synthesises established principles of causal inference into a practical decision sequence and does not represent a new theory of causation.

Causal Question → Treatment or Exposure → Assignment Mechanism → Counterfactual Strategy → Identification Assumptions → Alternative Explanations → Robustness Evidence → Defensible Causal Claim

1. Causal Question

The process begins by specifying the effect the study intends to estimate. Asking whether training is associated with employee productivity is different from asking whether introducing a particular training programme causes an increase in productivity. A precise causal question clarifies what the subsequent research design needs to identify.

2. Treatment or Exposure

The treatment must be defined precisely enough for treated and untreated conditions to have methodological meaning. Researchers should establish what intervention occurred, when it began, which units were exposed and whether treatment intensity or compliance varied. Ambiguity at this stage can make even technically sophisticated estimates difficult to interpret.

3. Assignment Mechanism

The researcher then investigates why some units received treatment while others did not. This step is especially important because the assignment mechanism often determines which quasi-experimental strategy is defensible. It can also reveal selection processes that threaten causal interpretation.

4. Counterfactual Strategy

The study must identify evidence capable of approximating what would have happened without treatment. Depending on the design, this might involve observations immediately below an eligibility threshold, an untreated group followed over the same period, a matched comparison group or an outcome trajectory established before the intervention. The relevant question is not whether a comparison exists, but whether it represents a credible counterfactual.

5. Identification Assumptions

Every quasi-experimental strategy requires assumptions connecting observed comparisons to the causal effect being estimated. These assumptions differ substantially across designs. Researchers should state them explicitly, explain why they are plausible in the particular research setting and examine observable implications where possible.

6. Alternative Explanations

The observed effect should be challenged by asking what else could have generated it. Selection, contemporaneous events, underlying trends, measurement changes, behavioural responses and other forms of confounding may provide competing explanations. The threats that matter depend on the research design and context rather than on a generic checklist.

7. Robustness Evidence

Design-appropriate diagnostics and robustness analyses can test whether conclusions depend excessively on particular specifications or assumptions. The purpose is not to run as many statistical models as possible, but to investigate vulnerabilities that could materially change the causal interpretation.

8. Defensible Causal Claim

The final conclusion should be calibrated to the evidence produced by the design. Researchers should not weaken credible quasi-experimental evidence merely because randomization was absent, but neither should they claim more certainty or generality than the identification strategy can support.

The framework can be summarised by one principle:

A quasi-experimental design becomes persuasive when it creates a credible counterfactual and provides evidence against plausible alternative explanations for the observed effect.

Threats to Causal Inference

Non-random treatment assignment makes selection a central concern in many quasi-experimental studies. Companies voluntarily adopting a new management system, for example, may already possess stronger leadership, more resources or greater willingness to change than non-adopters. If those characteristics also influence performance, a subsequent difference between adopters and non-adopters cannot automatically be attributed to the management system.

Changes occurring alongside the intervention create a different problem. Suppose employee productivity rises after a new incentive scheme, but the organisation introduces automation during the same period. The observed improvement may reflect the incentive scheme, automation or some combination of both. Similar difficulties arise when economic conditions, regulations, competitor behaviour or other external events affect the treatment and comparison groups differently.

Outcomes can also change naturally over time. Employee skills may improve through experience, sales may rise because a market is growing and extreme observations may move towards more typical values through regression toward the mean. Where treatment is introduced precisely because an outcome has become unusually poor, this last problem deserves particular attention because some apparent improvement may have occurred without intervention.

Measurement itself must remain sufficiently comparable. A change in performance indicators, survey instruments, data-recording systems or reporting practices can create an apparent treatment effect even when the underlying phenomenon has not changed correspondingly. Researchers should therefore identify the threats that are plausible for their specific design and show how the comparison strategy, evidence and analysis address them.

Application of Quasi-Experimental Research: an Example

Suppose a large retailer introduces a four-day working week across stores in one region while stores in another broadly comparable region continue operating under the existing five-day schedule. A researcher wants to estimate whether the policy reduces employee absenteeism.

Comparing the two regions after implementation would provide limited causal evidence because their absenteeism rates may already have differed. Comparing absenteeism only before and after the policy in the intervention region would also be insufficient because wider economic, seasonal or organisational changes could have affected attendance during the same period. The researcher therefore obtains monthly absenteeism data for both regions covering two years before implementation and one year afterwards.

A Difference-in-Differences strategy could compare the change in absenteeism in the intervention region with the contemporaneous change in the comparison region. Suppose absenteeism falls by three percentage points in the four-day-week region and by one percentage point in the comparison region. The causal analysis focuses on the differential change rather than treating the entire three-point decline as an effect of the policy.

The design still requires justification. The researcher would need to investigate whether pre-intervention patterns support the proposed comparison and whether other major changes occurred disproportionately in either region. If the four-day-week stores simultaneously received substantial additional staffing, for example, the resulting difference could not straightforwardly be attributed to the working-week policy. The strength of the study therefore comes from the logic of its counterfactual and the evidence supporting that logic, rather than from the Difference-in-Differences calculation alone.

Advantages and Limitations of Quasi-Experimental Research

Many consequential causal questions cannot realistically be investigated through randomized experiments. Governments rarely randomize major regulations, firms may not permit researchers to randomize strategic interventions, and ethical or practical constraints can make deliberate treatment assignment impossible. Quasi-experimental research allows researchers to exploit policies, thresholds, phased rollouts, institutional rules and other naturally occurring sources of variation to investigate causal effects under these conditions.

Research conducted around real organisational or policy interventions can also provide evidence in settings where decisions actually occur. A retailer testing a new operational system across selected locations, for example, may generate evidence with direct managerial relevance. Certain quasi-experimental designs can provide particularly strong causal leverage when their assignment mechanisms closely approximate the comparison that randomization would otherwise create.

The absence of random assignment nevertheless places substantial weight on assumptions that differ from one design to another. A sophisticated estimator cannot compensate for an implausible counterfactual strategy. Matching may produce excellent balance on measured characteristics while leaving important unmeasured confounding unresolved; Difference-in-Differences may be inappropriate when the comparison group cannot represent the untreated trajectory; an instrumental-variable estimate may be difficult to interpret if the proposed instrument does not satisfy the necessary identifying conditions.

The causal effect identified by a quasi-experiment may also apply to a narrower population than the researcher’s substantive question suggests. Regression Discontinuity commonly identifies effects around the assignment threshold, while instrumental-variable estimates can concern populations defined partly by their response to the instrument. Researchers therefore need to distinguish internal validity—whether the effect is credibly identified—from external validity—how far that effect can reasonably be generalised.

Practical data requirements create further constraints. Interrupted Time Series requires adequate observations across time, propensity-score approaches require sufficient overlap between treatment and comparison units, and RDD needs enough observations near the threshold. A design can be theoretically attractive yet unsuitable for the dataset available to a dissertation researcher.

Common Mistakes in Quasi-Experimental Research

Before-and-after change is frequently given more causal meaning than it can support. If sales increase after a marketing intervention, the timing alone does not establish that the intervention produced the increase. Seasonality, economic conditions, competitor behaviour and an existing growth trend could all generate change during the same period. A quasi-experimental argument needs a strategy for estimating what would have happened without treatment.

Convenient comparison groups can create an illusion of control without a credible counterfactual. Researchers sometimes select untreated organisations, regions or individuals simply because their data are accessible. The important issue is whether the comparison group represents the untreated outcome trajectory of the treated group sufficiently well for the intended causal inference.

Statistical sophistication is easily confused with design strength. Adding dozens of control variables or applying an advanced estimator may improve aspects of the analysis, but causal identification does not emerge automatically from model complexity. The assignment process and counterfactual strategy remain fundamental.

Method-specific assumptions can disappear behind software output. A statistically significant coefficient from RDD, IV, PSM or DiD does not demonstrate that the corresponding identification assumptions are credible. Researchers need to explain why the method fits the treatment-assignment process and provide relevant diagnostic or robustness evidence.

Propensity Score Matching can be described too casually as creating an experimental treatment and control group. Matching can improve comparability on observed covariates, but it does not randomize treatment and cannot automatically eliminate bias from unmeasured confounders. The resulting causal claim must reflect this limitation.

The label “quasi-experimental” should not substitute for an identification strategy. Two naturally occurring groups observed before and after an intervention do not become a persuasive quasi-experiment merely because the researcher uses that terminology. The methodology needs to show what creates causal leverage and why competing explanations are less consistent with the evidence.

Quasi-Experimental Research in Business Research

Business environments continually generate interventions that are implemented for operational rather than research purposes. Companies change prices, introduce technologies, redesign incentive systems, alter working arrangements and roll out new services across locations or customer groups. These decisions can create opportunities for quasi-experimental research when the implementation process generates a meaningful basis for comparison.

Consider a retailer introducing self-checkout technology first in stores above a particular transaction-volume threshold. A simple comparison between stores with and without self-checkout could be misleading because high-volume stores differ systematically from smaller locations. However, if implementation follows a sufficiently strict threshold rule, observations close to that cutoff may provide the basis for a Regression Discontinuity Design. Similarly, a phased policy rollout across regions might support a Difference-in-Differences strategy if the untreated regions provide an appropriate counterfactual.

This changes how researchers can think about business datasets. Instead of searching only for variables that can be entered into a regression model, they can examine how organisational decisions were made. Eligibility rules, thresholds, policy changes, geographic boundaries, phased implementation and external shocks may contain more causal information than the number of variables available in the dataset.

Quasi-experimental business research therefore combines substantive knowledge of the organisation with causal-design reasoning. Understanding why a policy affected one unit rather than another can be at least as important as selecting the statistical estimator used to analyse its effects.

Quasi-Experimental Research in the Age of AI and Digital Research

Digital platforms and organisational information systems increasingly provide the granular longitudinal data required for quasi-experimental research. Transaction histories, customer interactions, employee activity, pricing changes and policy implementation can sometimes be observed across thousands or millions of units. This makes it easier to investigate interventions at a level of temporal and behavioural detail that was previously difficult to obtain.

AI and machine-learning tools can support several stages of the analysis. They may help researchers identify potentially comparable observations, model complex relationships between covariates and outcomes, detect unusual patterns, automate diagnostic procedures and investigate heterogeneous treatment effects. These capabilities can be especially useful when conventional parametric models struggle with high-dimensional or nonlinear data.

Greater predictive capability, however, does not solve the counterfactual problem. An algorithm may predict extremely accurately which customers receive an intervention and what their subsequent behaviour will be without establishing what would have happened to those same customers under the alternative treatment condition. A model that predicts outcomes well is answering a different question from a design that identifies causal effects.

Generative AI creates a related methodological risk by making sophisticated analysis easier to execute than to understand. Researchers can increasingly produce matching procedures, regression models and robustness checks through natural-language instructions without fully understanding the assumptions behind them. The result may look technically impressive while lacking a defensible causal argument.

The most productive role for AI is therefore to support rather than replace identification reasoning. It can help challenge assumptions, explore alternative specifications, inspect balance, search for possible confounders and generate robustness analyses, while the researcher remains responsible for explaining why the assignment mechanism and counterfactual strategy support the causal claim.

When to Use Quasi-Experimental Research

Quasi-experimental research may be appropriate when:

• the research question concerns the causal effect of an intervention, treatment, policy or exposure;
• random assignment is unethical, infeasible or outside the researcher’s control;
• treatment assignment follows a rule, threshold, policy, timing process or other mechanism that may provide causal leverage;
• a credible comparison group or alternative counterfactual strategy can be constructed;
• repeated observations are available before and after an intervention;
• naturally occurring variation provides a defensible basis for causal identification;
• sufficient pre-treatment characteristics are available for an appropriate matching or propensity-score approach;
• the assumptions required by the proposed design can be clearly articulated and investigated;
• plausible competing explanations for the observed effect can be examined.

It is less appropriate when the research objective is purely descriptive or when the available treatment-assignment process and data provide no credible basis for constructing a counterfactual. In such circumstances, describing the study accurately as observational or associational is methodologically preferable to making a quasi-experimental causal claim that the design cannot support.

Dissertation Example

A dissertation titled “The Impact of a Minimum-Wage Increase on Employee Turnover in the UK Hospitality Sector” could use a quasi-experimental design if a policy change creates meaningful differences in exposure across firms, workers, regions or periods. The methodology chapter would begin by defining the causal effect of interest and explaining why random assignment is neither feasible nor under the researcher’s control.

The researcher would then describe the policy mechanism producing treatment exposure and justify the comparison used to approximate what would have happened to affected units without the minimum-wage change. If Difference-in-Differences were selected, the chapter would explain why the treatment and comparison groups provide a meaningful basis for comparison, define the pre- and post-policy periods and present evidence relevant to the design’s assumptions.

Potential competing explanations would form part of the methodological justification rather than appearing only as limitations at the end of the dissertation. Other labour-market reforms, regional economic changes or sector-specific shocks could affect employee turnover during the study period, so the researcher would explain how the design, data and robustness analyses address these possibilities and identify those that cannot be ruled out convincingly.

The methodology chapter would therefore present a coherent chain from causal question → treatment-assignment mechanism → counterfactual strategy → identifying assumptions → analysis → appropriately bounded causal conclusion. This demonstrates why the research is quasi-experimental and, more importantly, why the particular design provides evidence relevant to the causal claim.

Exam Tip

If you are asked why a study uses quasi-experimental research, saying “because random assignment was not possible” identifies the problem but does not explain the solution. A stronger methodological justification identifies what substitutes for randomization and why that alternative provides a credible counterfactual.

The answer will depend on the particular design. In Regression Discontinuity, the assignment threshold and comparability of observations around it are central. In Propensity Score Matching, the quality of measured pre-treatment covariates, overlap and achieved balance matter. In Difference-in-Differences, the credibility of the untreated counterfactual trajectory is fundamental.

A useful way to remember the logic is:

Absence of randomization defines the challenge; the quasi-experimental design explains how the researcher attempts to overcome it.

Need to determine whether your dissertation can support a causal research design—and which quasi-experimental method fits your research question and data?

Use Dudovskiy Research Assistant to develop a methodology tailored to your research topic, with clear justification for the research design, counterfactual strategy and causal claims.

References

Shadish, W.R., Cook, T.D. and Campbell, D.T. (2002) Experimental and Quasi-Experimental Designs for Generalized Causal Inference. Boston: Houghton Mifflin.

Holland, P.W. (1986) ‘Statistics and causal inference’, Journal of the American Statistical Association, 81(396), pp. 945–960.

Rubin, D.B. (2006) Matched Sampling for Causal Effects. Cambridge: Cambridge University Press.

Imbens, G.W. and Lemieux, T. (2008) ‘Regression discontinuity designs: A guide to practice’, Journal of Econometrics, 142(2), pp. 615–635.

Angrist, J.D. and Pischke, J.-S. (2009) Mostly Harmless Econometrics: An Empiricist’s Companion. Princeton, NJ: Princeton University Press.

Thoemmes, F. and Kim, E.S. (2011) ‘A systematic review of propensity score methods in the social sciences’, Multivariate Behavioral Research, 46(1), pp. 90–118.

Wing, C., Simon, K. and Bello-Gomez, R.A. (2018) ‘Designing difference in difference studies: Best practices for public health policy research’, Annual Review of Public Health, 39, pp. 453–469.

Bernal, J.L., Cummins, S. and Gasparrini, A. (2017) ‘Interrupted time series regression for the evaluation of public health interventions: a tutorial’, International Journal of Epidemiology, 46(1), pp. 348–355.