03 · What You Need to Know
Using a Hypothetical Target Trial to Structure Observational Causal Research
What Is a Hypothetical Target Trial?
A target trial is the randomized experiment that would ideally answer a specified causal question.
It defines the population, treatment strategies, assignment procedures, follow-up, outcomes, causal effect, and analysis before the researcher attempts to estimate the effect using observational evidence.
For example, a target trial might randomly assign eligible students at the beginning of a semester to receive a structured AI feedback intervention or conventional feedback.
The researcher would then measure writing performance after a specified period.
When such a trial cannot be conducted, its protocol can serve as a reference for designing an observational analysis.
Hernán and Robins (2016) describe how explicitly emulating a target trial can help avoid methodological problems in observational causal research.
Why Begin With a Trial That Will Not Actually Be Conducted?
Observational datasets are often organized according to administrative convenience rather than causal research requirements.
Records may contain exposure histories, outcomes, and participant characteristics without clearly defining when participants became eligible or when treatment strategies began.
Beginning with the hypothetical trial forces researchers to specify the intended intervention and comparison before selecting records or constructing exposure categories.
This can reveal whether the available data support the causal question.
It also makes design decisions more transparent, particularly when exposure and eligibility change over time.
The purpose is not to pretend randomization occurred. It is to make the observational design correspond as closely as possible to a coherent causal experiment while identifying differences that remain.
What Components Should a Target Trial Specify?
| Protocol component |
Question to specify |
Educational example |
| Eligibility criteria |
Who can enter the study, and when? |
Students enrolled in a defined writing course at semester start. |
| Treatment strategies |
What intervention alternatives are compared? |
Structured AI feedback versus conventional feedback. |
| Assignment |
How would eligible participants be allocated? |
Hypothetical random assignment to feedback strategies. |
| Time zero |
When do eligibility, assignment, and follow-up begin? |
The beginning of the instructional period. |
| Follow-up |
How long are outcomes observed? |
One academic semester. |
| Outcome |
What result is measured and when? |
Writing performance on a standardized final assessment. |
| Causal estimand |
What effect is the study intended to estimate? |
Effect of assignment to the feedback strategy among eligible students. |
| Analysis plan |
How would the target effect be estimated? |
Specified comparison and appropriate uncertainty estimation. |
These components should be coherent with one another. A treatment strategy that begins after the defined follow-up period, for example, cannot answer the intended causal question.
Why Are Eligibility Criteria Important?
Eligibility criteria define the population to which the target causal effect applies.
For example, a study may include students who are enrolled in a writing course, have not yet received the intervention, and are eligible for either feedback strategy at the beginning of the semester.
These conditions should be determined using information available at the relevant eligibility time.
If researchers select participants according to events that occur later, the resulting sample may be affected by post-baseline selection.
Eligibility should therefore be specified before exposure histories and outcomes are analyzed.
What Are Treatment Strategies?
Treatment strategies describe the interventions whose causal effects are being compared.
For example, one strategy might involve offering students access to a structured AI feedback system, while another provides conventional teacher feedback.
A different question might compare sustained weekly use of AI feedback with no AI feedback use throughout the semester.
These strategies are not equivalent.
The first concerns assignment or access, while the second concerns sustained behavior that may depend on evolving student characteristics.
Researchers should define treatment strategies precisely enough that the relevant potential outcomes are meaningful.
Ambiguous categories such as AI users and nonusers may conceal multiple intervention versions.
What Is Time Zero?
Time zero is the beginning of follow-up for the causal comparison, aligned with eligibility and the assignment or initiation of the treatment strategies.
For example, if the target trial would assign students to feedback strategies at the beginning of the semester, follow-up should begin at that corresponding point.
Defining time zero consistently helps prevent differences in the opportunity to experience outcomes.
It also prevents researchers from classifying participants according to future exposure information without accounting for the period needed to acquire that classification.
Time-zero alignment is one of the most consequential contributions of the target trial framework.
What Is Immortal Time Bias?
Immortal time bias can arise when a period during which a participant must remain event-free or under observation to qualify for an exposure category is handled incorrectly.
Suppose researchers classify students as sustained AI users only if they complete at least ten weeks of system use.
Students must remain enrolled and observed long enough to satisfy that definition.
If follow-up begins at semester start but sustained users are identified using later information, the comparison may give that group a structurally different opportunity to experience the outcome or withdrawal event.
Appropriate treatment strategies, time alignment, and analytical methods are needed to avoid this problem.
Hernán and colleagues (2016) emphasize how specifying the target trial can reveal biases arising from inappropriate time-related design decisions.
Why Should Eligibility, Assignment, and Follow-Up Be Aligned?
These events should correspond to the same meaningful starting point in the causal design.
If one group begins follow-up when treatment starts while another begins months earlier, differences may arise from unequal observation periods rather than treatment effects.
Similarly, using future information to define eligibility may create selection bias.
Researchers should specify when participants become eligible, when intervention strategies are assigned or initiated, and when outcomes begin to be observed.
The observational analysis should then approximate this structure as closely as the available data permit.
Does a Target Trial Require Actual Randomization?
No. The hypothetical target trial specifies random assignment as part of the ideal experiment, but the observational study does not claim that randomization occurred.
Instead, the researcher attempts to estimate the target effect using nonrandomized exposure information.
This requires assumptions about treatment selection and potential outcomes.
Conditional exchangeability, positivity, and consistency are particularly relevant in many observational causal analyses.
The target trial framework clarifies these assumptions rather than eliminating them.
What Is the Difference Between a Target Trial and Target Trial Emulation?
A target trial is the hypothetical randomized experiment defining the causal question.
Target trial emulation is the process of designing and analyzing observational data to approximate that trial's protocol.
For example, the researcher may specify an ideal randomized comparison at semester start and then identify observational records that satisfy corresponding eligibility, treatment, and follow-up definitions.
The distinction matters because specifying a hypothetical trial is a conceptual design step, while emulation requires operational decisions about real data.
The detailed procedures of target trial emulation depend on the treatment strategies, available information, and causal estimand.
What Is the Role of Conditional Exchangeability?
Conditional exchangeability requires that the observational comparison adequately account for common causes of treatment and outcome through an appropriate covariate set.
For example, students' prior writing ability, digital access, and motivation may influence both AI feedback adoption and later performance.
If important confounders are unmeasured, the observational estimate may remain biased.
Specifying the target trial helps identify what should have been measured before treatment assignment.
However, the framework cannot recover variables that were never recorded.
Researchers should distinguish a clearly specified target trial from a successfully identified causal effect.
Why Is Positivity Important?
Positivity requires that the treatment strategies being compared are possible across relevant covariate profiles in the target population.
Suppose students without reliable internet access cannot realistically use the AI feedback system.
The available data may contain no comparable exposed observations for those students.
Estimating the effect of sustained AI use for that subgroup may therefore require unsupported extrapolation.
Researchers may need to revise eligibility criteria, redefine the intervention, or restrict the target population.
The hypothetical trial should describe a comparison that is meaningful and feasible for the population under consideration.
What Is Consistency in a Target Trial Framework?
Consistency connects observed outcomes with the potential outcomes under the treatment strategies actually followed.
For example, if AI-supported feedback includes several substantially different forms of use, a single exposure label may not correspond to one well-defined intervention strategy.
Researchers should specify the intervention sufficiently clearly to support interpretation.
However, treatment strategies may legitimately include flexible or adaptive components when these are explicitly defined.
The important issue is whether the exposure definition corresponds to a meaningful causal intervention.
What Is the Difference Between an Assignment Effect and a Per-Protocol Effect?
An assignment effect concerns outcomes under alternative treatment assignments or offers.
A per-protocol effect concerns outcomes under specified adherence to intervention strategies.
For example, offering access to AI feedback may have a different effect from ensuring sustained use throughout the semester.
The observational data requirements and identifying assumptions may differ substantially between these questions.
Researchers should define the estimand before constructing exposure groups.
Hernán and Robins (2017) discuss the distinction between intention-to-treat and per-protocol effects and the challenges of estimating adherence-related causal effects.
Can a Target Trial Framework Address Time-Varying Treatments?
Yes. It can clarify intervention strategies that change over time.
For example, a strategy may require students to complete a specified feedback activity each week rather than simply use the system at least once.
However, treatment decisions may depend on evolving performance, confidence, or workload.
Some of these characteristics may be affected by earlier treatment.
Conventional adjustment may be inadequate when time-varying confounders are also consequences of previous exposure.
Appropriate longitudinal causal methods may be needed under defensible assumptions.
The target trial framework clarifies the question but does not automatically determine the correct estimator.
Can the Same Participant Enter Multiple Target Trials?
In some longitudinal observational settings, researchers may emulate a sequence of trials beginning at different eligible time points.
For example, students may become eligible to initiate a feedback intervention at several points during an academic program.
Each eligible time may define a separate trial-like comparison under a specified protocol.
However, repeated eligibility creates dependencies because the same participant may contribute to multiple emulated trials.
Analyses must account appropriately for this structure.
Sequential trial emulation is not necessary for every study, but it can be useful when treatment initiation decisions occur repeatedly.
What If Treatment Strategies Cannot Be Distinguished in the Available Data?
The observational dataset may not contain sufficient information to identify the intended intervention conditions.
For example, platform records may show login frequency without indicating whether students received meaningful feedback or applied it to revisions.
If the target trial concerns sustained formative feedback engagement, login counts may be an inadequate exposure measure.
Researchers should evaluate whether the available records support the intervention definition.
Where they do not, the causal question may need to be revised or additional data collected.
The target trial framework is valuable partly because it exposes such limitations before researchers interpret an association as a treatment effect.
When Is the Target Trial Framework Particularly Useful?
It is especially useful when researchers investigate well-defined interventions using longitudinal observational data.
Examples include treatment initiation, sustained intervention use, switching between strategies, and policy implementation with clearly defined eligibility and follow-up.
The framework may also help clarify comparisons involving institutional interventions when the causal contrast can be specified coherently.
However, the approach should not be applied mechanically to every observational study.
A descriptive survey estimating AI use prevalence does not require a hypothetical randomized trial because its objective is not causal intervention estimation.
Likewise, an exploratory qualitative study may investigate experiences without defining treatment strategies or potential outcomes.
Can a Target Trial Be Designed Around an Impossible Intervention?
A hypothetical trial should represent a sufficiently well-defined and meaningful intervention contrast.
For example, randomly assigning students to receive structured feedback is conceptually clearer than assigning them to possess an unspecified level of motivation.
Some exposures cannot be directly manipulated in the way a treatment can, and causal questions involving them require additional conceptual care.
Researchers should distinguish an intervention on a modifiable process from an ill-defined attempt to manipulate a fixed characteristic.
The target trial should clarify the intervention rather than conceal ambiguity behind experimental terminology.
Does Target Trial Emulation Eliminate Confounding?
No. Emulation improves design clarity and may prevent avoidable biases, but it does not reproduce random assignment.
Students who initiate AI feedback may differ systematically from those who do not.
Appropriate adjustment may address selected measured confounding under suitable assumptions, but unmeasured confounding may remain.
The question of whether adjustment establishes causation therefore remains central.
Researchers should not describe an observational study as randomized merely because it was organized according to a target trial protocol.
How Does Target Trial Design Relate to Alternative Explanations?
Specifying a hypothetical trial can reveal how an observational comparison might be biased.
For example, defining exposure groups using future usage information may introduce time-related selection bias.
Allowing different eligibility dates across groups may create unequal follow-up.
Omitting baseline achievement may leave important confounding unaddressed.
Researchers should use the target trial protocol to identify where the observational analysis departs from the ideal experiment and which assumptions are needed to justify those departures.
The broader principle of designing causal studies around alternative explanations is therefore closely related to target trial specification.
What Should Be Reported in an Observational Target Trial Study?
Researchers should describe the hypothetical target trial and explain how each protocol component was implemented using observational data.
Eligibility criteria, treatment strategies, time zero, follow-up, outcomes, estimands, and analytical procedures should be explicit.
Departures from the hypothetical randomized trial should be identified.
The methodology should justify confounder selection, explain exposure measurement, and evaluate relevant positivity, consistency, and missing-data assumptions.
Researchers should distinguish the design specification from evidence that the causal effect has been identified successfully.
Transparency is particularly important when the available data were collected for administrative purposes rather than the causal question being investigated.
Watch Out
Do not assume that calling an observational analysis a target trial emulation makes it equivalent to a randomized trial. The framework improves causal question specification and design alignment, but the resulting estimate still depends on observational identification assumptions and the quality of available data.
