Interrupted Time Series Analysis in Ecology
By Dr. Zubair Khalid, DVM, MS, PhD ·
Quick Answer
- Interrupted time series analysis evaluates whether an intervention changed an ecological outcome by comparing pre-intervention and post-intervention trends in longitudinal data.
- Design the study with a clearly defined intervention point, sufficient pre-intervention and post-intervention observations, and a control series when feasible.
- The method assumes the pre-intervention trend would have continued without the intervention, and this counterfactual assumption cannot be directly verified.
What Is Interrupted Time Series Analysis in Ecology
Interrupted time series analysis is a quasi-experimental design used to evaluate the effect of an intervention on an outcome measured repeatedly over time. In ecology, researchers apply this method to assess whether a conservation policy, a management action, a pollution control, or a species introduction changed the level or slope of an ecological indicator. The design compares the observed post-intervention trajectory with the trajectory projected from the pre-intervention data.
The term interrupted refers to the structure of the data. A continuous series of measurements is split at a known point in time when the intervention occurred. The analysis estimates whether the post-interruption segment differs from what the pre-interruption segment would have predicted. This approach is valuable in ecology because randomized controlled trials are often impractical for landscape-scale or policy-level interventions.
The design belongs to the broader category of quasi-experimental methods. Unlike a true experiment, the researcher does not randomly assign units to treatment and control conditions. Instead, the analysis relies on temporal structure and statistical modeling to estimate the intervention effect. The National Library of Medicine provides access to research methods references that describe the principles of quasi-experimental designs and their application in health and environmental research.
Core Principles of Interrupted Time Series Design
The Counterfactual Logic
The central logic of interrupted time series analysis rests on a counterfactual question. What would the ecological outcome have been if the intervention had not occurred? The pre-intervention segment of the series provides the basis for projecting this counterfactual trajectory. The difference between the projected trajectory and the observed post-intervention trajectory is the estimated intervention effect.
This logic requires that the pre-intervention trend is stable enough to serve as a valid projection baseline. If the pre-intervention series shows strong seasonal cycles, irregular fluctuations, or a changing slope, the projection becomes less reliable. The researcher must document the stability of the pre-intervention series and consider whether the projection is defensible.
Segmentation of the Time Series
The time series is divided into two segments at the intervention point. The pre-intervention segment establishes the baseline level and trend. The post-intervention segment captures the outcome after the intervention begins. The analysis estimates four parameters of interest. The baseline level is the expected outcome at the start of the series. The baseline trend is the slope of the pre-intervention segment. The level change is the immediate shift in the outcome at the intervention point. The trend change is the difference between the pre-intervention slope and the post-intervention slope.
These parameters allow the researcher to distinguish between an abrupt intervention effect and a gradual effect. An immediate level change suggests the intervention had a rapid impact. A trend change suggests the intervention altered the trajectory over time. Both effects can occur together.
The Role of Autocorrelation
Ecological time series data are often autocorrelated. The value at one time point is correlated with values at nearby time points. This autocorrelation violates the independence assumption of ordinary least squares regression. If ignored, the analysis can produce standard errors that are too small, leading to false conclusions about the significance of the intervention effect.
The researcher must test for autocorrelation in the residuals of the fitted model. Common diagnostic tools include the Durbin-Watson statistic and inspection of residual autocorrelation plots. When autocorrelation is present, the model must be adjusted. Options include adding autoregressive terms, using generalized least squares, or applying a Newey-West correction to the standard errors.
Core Principles of Interrupted Time Series Analysis
The Intervention Point Must Be Known
The researcher must know the exact time at which the intervention occurred. This point is the interruption in the series. If the intervention was implemented gradually or if the timing is uncertain, the analysis becomes more difficult. The researcher should document the intervention date and the mechanism by which the intervention was expected to affect the ecological outcome.
The Outcome Must Be Measured Consistently
The outcome variable must be measured in the same way before and after the intervention. If the measurement method changes at the same time as the intervention, the analysis cannot separate the intervention effect from the measurement change. The researcher should verify that the data collection protocol remained constant across the study period.
The Pre-Intervention Series Must Be Sufficient
A reliable baseline projection requires enough pre-intervention data points. Short baselines produce unstable trend estimates. The number of data points needed depends on the variability of the outcome and the frequency of measurement. A general principle is that the pre-intervention series should be long enough to establish a stable trend. The researcher should examine the pre-intervention data for outliers, missing values, and structural changes.
The Post-Intervention Series Must Be Sufficient
The post-intervention segment must also be long enough to detect the intervention effect. If the effect is gradual, a short post-intervention series may not capture the full trajectory. The researcher should consider the expected time course of the intervention effect when planning the analysis.
At a Glance
| Design Element | What the Researcher Must Decide | Common Approach in Ecology |
|---|---|---|
| Intervention point | The exact time the intervention began | Use a documented policy date, management action date, or event date |
| Outcome variable | The ecological measure repeated over time | Species abundance, water quality, habitat area, or population growth rate |
| Pre-intervention length | Number of data points before the intervention | At least 8 to 12 points when data are collected monthly or quarterly |
| Post-intervention length | Number of data points after the intervention | At least 8 to 12 points to detect gradual effects |
| Autocorrelation handling | Whether residuals are independent | Test with Durbin-Watson or residual plots, then adjust the model |
| Control series | A comparison series without the intervention | A nearby unmanaged site or an unaffected species |
Practical Workflow for an Interrupted Time Series Study
Step 1 Define the Research Question
The research question must specify the intervention, the ecological outcome, and the time frame. A clear question might be whether a new fishing regulation changed the abundance of a target fish species. The question should also specify the expected effect. Is the effect expected to be immediate or gradual? This expectation guides the interpretation of the level change and trend change parameters.
Step 2 Assemble the Time Series Data
The researcher must compile the outcome data over the full study period. The data should be recorded at regular intervals. Irregular intervals complicate the analysis because the model assumes equally spaced time points. If the data are irregular, the researcher must either aggregate the data to a common interval or use a model that accounts for the spacing.
Step 3 Document the Intervention
The researcher must document the intervention date and the expected mechanism of action. This documentation is essential for the analysis and for the interpretation of the results. The researcher should also note any other events that occurred around the same time and could have affected the outcome. These confounding events are a major threat to the validity of the design.
Step 4 Plot the Data
A visual inspection of the time series is the first step in the analysis. The plot should show the pre-intervention and post-intervention segments. The researcher should look for an obvious change in level or trend at the intervention point. The plot also reveals outliers, missing data, and unusual patterns that could affect the analysis.
Step 5 Fit the Segmented Regression Model
The core of the analysis is a segmented regression model. The model includes a time variable, an indicator variable for the post-intervention period, and an interaction term between time and the post-intervention indicator. The time variable captures the pre-intervention trend. The indicator variable captures the immediate level change. The interaction term captures the change in trend after the intervention.
The model can be written as follows. The outcome at time t is equal to the baseline level plus the baseline trend multiplied by time, plus the level change multiplied by the post-intervention indicator, plus the trend change multiplied by the interaction term, plus the error term. The error term must be examined for autocorrelation.
Step 6 Test for Autocorrelation
After fitting the model, the researcher must examine the residuals. The Durbin-Watson statistic tests for first-order autocorrelation. A value near 2 indicates no autocorrelation. Values below 1.5 or above 2.5 suggest autocorrelation. The researcher should also plot the residuals against time and inspect the autocorrelation function.
If autocorrelation is present, the model must be adjusted. The researcher can add autoregressive terms to the model, use generalized least squares, or use a robust variance estimator. The choice of adjustment should be based on the structure of the autocorrelation.
Step 7 Interpret the Results
The model produces estimates for the baseline level, baseline trend, level change, and trend change. The researcher must interpret these estimates in the context of the research question. A significant level change indicates an immediate effect. A significant trend change indicates a gradual effect. The researcher should also report the confidence intervals for the estimates.
Step 8 Assess the Validity of the Design
The researcher must consider the threats to the validity of the design. The main threat is the possibility that the pre-intervention trend would not have continued without the intervention. The researcher should examine the pre-intervention data for stability and consider whether any other events could explain the observed change. The researcher should also consider whether the outcome measurement was consistent across the study period.
Options and Tradeoffs in Interrupted Time Series Analysis
Segmented Regression
Segmented regression is the most common approach to interrupted time series analysis. The model is simple to fit and the results are easy to interpret. The main limitation is that the model assumes a linear trend in each segment. If the ecological outcome follows a nonlinear trajectory, the model may not fit the data well.
Autoregressive Integrated Moving Average Models
An autoregressive integrated moving average model can be used to model the time series structure more explicitly. This approach is useful when the data show strong seasonal patterns or complex autocorrelation. The model is more complex than segmented regression and requires more statistical expertise. The interpretation of the intervention effect is also less direct.
Generalized Additive Models
A generalized additive model can be used when the relationship between time and the outcome is nonlinear. The model allows the trend to vary smoothly over time. The intervention effect is estimated as the difference between the observed trajectory and the projected trajectory. This approach is flexible but requires careful specification of the smoothing parameters.
Control Series
A control series is a second time series that is not exposed to the intervention. The control series can be used to strengthen the design. If the control series shows no change at the intervention point, the researcher can be more confident that the observed change in the intervention series is due to the intervention. The control series should be as similar as possible to the intervention series, except for the intervention.
Observations and Measurements
Data Quality Checks
The researcher must verify the quality of the time series data before analysis. The data should be checked for missing values, outliers, and measurement errors. Missing values can be handled by interpolation or by using a model that accommodates missing data. Outliers should be examined to determine whether they are data errors or genuine ecological events.
Measurement Consistency
The outcome must be measured consistently across the study period. If the measurement method changed, the analysis cannot distinguish the intervention effect from the measurement change. The researcher should document the measurement protocol and verify that it remained constant.
Temporal Resolution
The temporal resolution of the data affects the analysis. Data collected at a fine temporal resolution, such as daily or weekly, provide more data points and more statistical power. Data collected at a coarse resolution, such as annually, provide fewer data points and less power. The researcher should consider the temporal resolution when planning the study.
Records and Documentation
Study Protocol
The researcher should document the study protocol before the analysis begins. The protocol should specify the research question, the intervention point, the outcome variable, the data source, and the planned analysis. The protocol should also specify the criteria for interpreting the results.
Data Management
The data should be stored in a structured format with clear variable names and documentation. The data management plan should describe how the data were collected, cleaned, and stored. The National Institutes of Health Data Management and Sharing Policy describes the expectations for data management and sharing in NIH-funded research. The researcher should follow the data management requirements of their funding agency.
Analysis Code
The analysis code should be documented and stored with the data. The code should be clear and reproducible. The researcher should also document the software version and the statistical methods used. This documentation allows other researchers to reproduce the analysis.
Common Failure Patterns
Too Few Pre-Intervention Data Points
A common failure is to have too few pre-intervention data points to establish a stable baseline trend. The projected counterfactual is then unreliable, and the intervention effect estimate is uncertain. The researcher should plan the study to include a sufficient pre-intervention period.
Ignoring Autocorrelation
Another common failure is to ignore autocorrelation in the residuals. The standard errors are then too small, and the analysis may produce false positive results. The researcher must test for autocorrelation and adjust the model when necessary.
Confounding Events
A third common failure is to ignore other events that occurred at the same time as the intervention. These events can produce a change in the outcome that is incorrectly attributed to the intervention. The researcher should identify potential confounding events and consider their effect on the results.
Changing Measurement Methods
A fourth common failure is to change the measurement method at the same time as the intervention. The analysis cannot separate the intervention effect from the measurement change. The researcher should verify that the measurement method was constant across the study period.
Limitations of Interrupted Time Series Analysis
The Counterfactual Assumption
The design relies on the assumption that the pre-intervention trend would have continued without the intervention. This assumption cannot be verified directly. The researcher can only assess the plausibility of the assumption by examining the pre-intervention data and considering other events.
The Need for a Known Intervention Point
The design requires a known intervention point. If the intervention was implemented gradually or if the timing is uncertain, the analysis is more difficult. The researcher should document the intervention date and the implementation process.
The Need for Consistent Measurement
The design requires consistent measurement across the study period. If the measurement method changed, the analysis is invalid. The researcher should verify the measurement consistency.
The Need for Sufficient Data
The design requires sufficient data points before and after the intervention. The number of data points needed depends on the variability of the data and the expected effect size. The researcher should plan the study to include a sufficient number of data points.
Safety and Regulatory Context
Interrupted time series analysis is a statistical method and does not involve direct manipulation of ecological systems. The safety context is therefore limited to the data collection activities that produce the time series. The researcher should follow the safety protocols of their institution for field data collection and laboratory analysis.
The researcher should also follow the publication ethics standards of their field. The Committee on Publication Ethics provides core practices for authorship, peer review, data, conflicts, and misconduct. The researcher should ensure that the study is reported transparently and that the data are shared in accordance with the expectations of the journal and the funding agency.
Professional Escalation Criteria
The researcher should seek additional statistical expertise when the data show complex autocorrelation, when the pre-intervention trend is unstable, or when the intervention point is uncertain. The researcher should also seek expertise when the analysis produces results that are inconsistent with the ecological understanding of the system. The researcher should consult with a statistician or a colleague with experience in time series analysis.
Decision Framework for Selecting Between Interrupted Time Series and Alternative Designs
When Interrupted Time Series Is the Appropriate Choice
The decision to use interrupted time series analysis should be made deliberately instead of by default. Ecologists often face a choice among several quasi-experimental designs, and the selection depends on the structure of the available data and the nature of the intervention. Interrupted time series is the strongest option when the intervention occurred at a known point in time, the outcome was measured repeatedly before and after that point, and no randomized comparison group exists.
The design is particularly well suited to policy evaluations where the intervention applies to an entire population or landscape. A fishing regulation that applies to all vessels in a management area, a protected area designation that changes land use rules, or a pollution standard that affects all discharges in a watershed are examples where individual-level randomization is impossible. In these cases, the temporal structure of the data provides the only source of comparison.
The researcher should confirm that the intervention point is a single event instead of a gradual process. If the intervention was phased in over several years, the analysis becomes more complicated. The researcher must then decide whether to treat the start of the phase-in as the intervention point or to model the phase-in period separately. The National Library of Medicine provides access to research methods references that describe the conditions under which quasi-experimental designs are appropriate and the assumptions they require.
When a Control Series Design Is Preferable
A control series design should be selected when a comparable unexposed series is available. The control series does not replace the interrupted time series analysis. It supplements it. The researcher fits the same segmented regression model to the control series and examines whether the control series shows a level change or trend change at the intervention point. If the control series shows no change, the evidence for an intervention effect is strengthened.
The control series must be selected with care. The ideal control series is measured with the same protocol, at the same temporal resolution, and over the same time period as the intervention series. The control series should be ecologically similar to the intervention series. A nearby unmanaged site, a similar watershed without the policy, or a species not targeted by the intervention are examples. The researcher should document the reasons the control series is expected to respond to the same external factors as the intervention series.
The control series is most valuable when the researcher suspects that external factors could have caused the observed change. If a regional drought occurred at the same time as the intervention, both the intervention series and the control series would show a change. The comparison allows the researcher to separate the intervention effect from the drought effect. Without the control series, the drought is a plausible alternative explanation.
When a Before-After Comparison Is Insufficient
A simple before-after comparison, which compares the mean outcome before the intervention with the mean outcome after the intervention, is insufficient when the outcome shows a pre-existing trend. If the outcome was already increasing before the intervention, a before-after comparison will attribute the ongoing increase to the intervention. The interrupted time series design avoids this error by modeling the pre-intervention trend and comparing the post-intervention trajectory with the projection of that trend.
The researcher should use a before-after comparison only when the pre-intervention series is flat and stable. This situation is rare in ecology because most ecological indicators show seasonal cycles, gradual trends, or responses to environmental drivers. The interrupted time series design is the safer choice whenever the pre-intervention series shows any systematic pattern.
When a Randomized Design Is Feasible
Randomized designs are preferred when they are feasible. If the intervention can be assigned randomly to some sites or some time periods and not others, the randomized design provides a stronger basis for causal inference. The researcher should consider whether randomization is possible before defaulting to an interrupted time series design.
Randomized designs are often feasible for smaller-scale interventions. A researcher testing a restoration technique can randomly assign plots to treatment and control. A researcher testing a new monitoring protocol can randomly assign sampling locations. In these cases, the randomized design avoids the counterfactual assumption that is central to interrupted time series analysis.
The interrupted time series design remains valuable even when randomization is possible. The design can be used to evaluate the long-term effect of an intervention after the randomized trial ends. The design can also be used when the randomized trial is not representative of the broader ecological system.
Decision Criteria Summary
The researcher should select the interrupted time series design when all of the following conditions hold. The intervention occurred at a known point in time. The outcome was measured at regular intervals before and after the intervention. The pre-intervention series is long enough to establish a stable trend. The measurement protocol remained constant across the study period. No randomized comparison group is available.
The researcher should consider a control series when a similar unexposed series is available. The researcher should consider a randomized design when the intervention can be assigned to units. The researcher should consider a before-after comparison only when the pre-intervention series is flat.
Practical Implementation Steps for the Decision Framework
Step 1 Inventory the Available Data
The researcher begins by listing the outcome data that are available. The list should include the time period covered, the temporal resolution, the number of data points before and after the intervention, and the measurement protocol. The researcher should also list any other time series that could serve as a control series.
The inventory should be written down and reviewed before the analysis begins. The inventory helps the researcher identify whether the data are sufficient for an interrupted time series design. The inventory also helps the researcher identify whether a control series is available.
Step 2 Assess the Intervention Structure
The researcher then assesses the structure of the intervention. The intervention should have a single identifiable start date. The researcher should document the date and the mechanism by which the intervention was expected to affect the outcome. If the intervention was implemented gradually, the researcher must decide whether the start date or the completion date is the appropriate intervention point.
Step 3 Evaluate the Pre-Intervention Series
The researcher plots the pre-intervention data and examines the trend. The pre-intervention series should show a stable trend. If the pre-intervention series shows a changing slope, strong seasonality, or large fluctuations, the projection of the counterfactual is less reliable. The researcher should consider whether the pre-intervention series is long enough to establish the trend.
Step 4 Check for a Control Series
The researcher identifies whether a control series is available. The control series should be measured with the same protocol over the same time period. The control series should be ecologically similar to the intervention series. If a control series is available, the researcher should plan to fit the same model to the control series.
Step 5 Document the Decision
The researcher should document the reasons for selecting the interrupted time series design. The documentation should include the inventory of the data, the assessment of the intervention point, and the assessment of the pre-intervention series. The documentation should also include the reasons for selecting or not selecting a control series. This documentation is part of the study protocol and supports the transparency of the analysis.
Records and Measurements for the Decision Framework
Intervention Documentation Record
The researcher should maintain a record of the intervention. The record should include the intervention date, the intervention type, the expected mechanism of action, and any other events that occurred at the same time. The record should also include the source of the information about the intervention date. This record is essential for the analysis and for the interpretation of the results.
Data Inventory Record
The researcher should maintain a record of the data that are available for the analysis. The record should include the outcome variable, the temporal resolution, the time period covered, the number of data points before and after the intervention, and the measurement protocol. The record should also include the source of the data and any known data quality issues.
Control Series Assessment Record
The researcher should maintain a record of the assessment of the control series. The record should include the control series, the reasons for selecting the control series, and the assessment of the similarity between the control series and the intervention series. The record should also include the results of the control series analysis.
Decision Log
The researcher should maintain a decision log that records the key decisions made during the design of the study. The log should include the decision to use the interrupted time series design, the decision about the intervention point, the decision about the control series, and the decision about the analysis model. The log should also include the reasons for each decision.
Common Failure Patterns in Design Selection
Selecting the Design Without a Data Inventory
A common failure is to select the interrupted time series design without first inventorying the available data. The researcher may later discover that the pre-intervention series is too short or that the measurement protocol changed at the intervention point. The researcher should complete the data inventory before the design is selected.
Ignoring the Availability of a Control Series
A second common failure is to ignore the availability of a control series. The researcher may have access to a nearby unmanaged site or an unaffected species but does not include it in the analysis. The control series can strengthen the design and should be used when available.
Selecting the Design When the Intervention Is Gradual
A third common failure is to select the interrupted time series design when the intervention was implemented gradually. The researcher must identify a single intervention point, but the gradual implementation makes the interpretation of the results difficult. The researcher should consider whether the intervention can be treated as a single event or whether a different design is needed.
Using a Before-After Comparison When a Trend Exists
A fourth common failure is to use a before-after comparison when the pre-intervention series shows a trend. The before-after comparison will attribute the pre-existing trend to the intervention. The researcher should use the interrupted time series design whenever the pre-intervention series shows any systematic pattern.
Comparison of Design Options
| Design Option | Data Requirement | Strength of Causal Evidence | Main Limitation |
|---|---|---|---|
| Interrupted time series | Single outcome series with pre and post data | Moderate | Counterfactual assumption |
| Interrupted time series with control series | Intervention series plus a similar unexposed series | Stronger | Requires a valid control series |
| Before-after comparison | Outcome measured before and after the intervention | Weak | Cannot separate the intervention effect from a pre-existing trend |
| Randomized design | Units randomly assigned to treatment and control | Strongest | Often infeasible for policy-level interventions |
Professional Escalation Criteria for Design Selection
The researcher should seek additional statistical expertise when the intervention point is uncertain. The researcher should also seek expertise when the pre-intervention series is short or unstable. The researcher should seek expertise when the control series is not clearly similar to the intervention series. The researcher should also seek expertise when the intervention was implemented gradually and the analysis requires a more complex model.
The researcher should consult with a statistician or a colleague with experience in quasi-experimental design. The consultation should occur before the analysis is finalized. The consultation should address the selection of the design, the identification of the intervention point, and the handling of the control series.
Reporting the Design Decision
The researcher should report the design decision in the methods section of the study. The report should describe the data inventory, the intervention point, the pre-intervention series, and the control series. The report should also describe the reasons for the design selection. The EQUATOR Network provides a collection of reporting guidelines that describe the reporting requirements for quasi-experimental studies. The researcher should select the appropriate guideline and follow the reporting requirements.
The researcher should also report the design decision in the study protocol. The protocol should be documented before the analysis begins. The protocol should specify the design, the intervention point, the outcome variable, and the planned analysis. The protocol should also specify the criteria for interpreting the results. The National Institutes of Health Data Management and Sharing Policy describes the expectations for data management and sharing in NIH-funded research. The researcher should follow the data management requirements of their funding agency.
Frequently Asked Questions
What is the minimum number of data points needed for an interrupted time series analysis?
The number of data points needed depends on the variability of the data and the expected effect size. A general rule is to have at least 8 to 12 points before and after the intervention. More points provide more stable estimates and more statistical power.
How do I handle autocorrelation in my time series data?
Test the residuals of the model for autocorrelation using the Durbin-Watson statistic or residual plots. If autocorrelation is present, adjust the model by adding autoregressive terms, using generalized least squares, or using a robust model.
What is the difference between a level change and a trend change?
A level change is an immediate shift in the outcome at the intervention point. A trend change is a change in the slope of the outcome after the intervention. Both effects can be estimated in the segmented regression model.
Can I use an interrupted time series design if the intervention was implemented gradually?
The design is more difficult when the intervention is gradual. The researcher must define the intervention point and consider whether the gradual implementation affects the interpretation of the results. The analysis may need to be adjusted to account for the gradual implementation.
What is a control series and why is it useful?
A control series is a second time series that is not exposed to the intervention. The control series provides a comparison for the intervention series. If the control series shows no change at the intervention point, the researcher can be more confident that the change in the intervention series is due to the intervention.
How do I report the results of an interrupted time series analysis?
Report the estimated level change and trend change with confidence intervals. Describe the model used, the autocorrelation adjustment, and the validity of the design. The reporting should be transparent and follow the reporting guidelines of the journal.
What are the main threats to the validity of an interrupted time series design?
The main threats are the counterfactual assumption, confounding events, and measurement changes. The researcher should examine the pre-intervention data, identify potential confounding events, and verify the measurement consistency.
Where can I find reporting guidelines for my study?
The EQUATOR Network provides a collection of reporting guidelines for health and natural science research. The researcher should select the appropriate guideline for the study design and follow the reporting requirements.
Related Diagnostic Guides
- Cell Membrane Proteins: Structure, Function, and Experimental Analysis
- Microscopy Image Analysis: Software and Workflows for Quantitative Data
- Cell Signaling Pathways: A Primer for Interpreting Experimental Data
References and Further Reading
- Research Methods Resources. National Library of Medicine.
- EQUATOR Network. EQUATOR Network.
- Core Practices. Committee on Publication Ethics.
- NIH Grants and Funding. National Institutes of Health.
- ORCID for Researchers. ORCID.
- Data Management and Sharing Policy. National Institutes of Health.
- Laboratory Biosafety Manual. World Health Organization.
- NCBI Literature Resources. National Center for Biotechnology Information.
- Segmented regression analysis of interrupted time series studies in medication use research.. Journal of clinical pharmacy and therapeutics, 2002.
- Use of interrupted time series analysis in evaluating health care quality improvements.. Academic pediatrics, 2013.
- Experimental and quasi-experimental designs in implementation research.. Psychiatry research, 2020.
This article is educational and does not replace validated laboratory procedures, institutional biosafety review, manufacturer instructions, or professional interpretation.