RESEARCH DISCIPLINE
Government Administrative Data Analysis
Administrative Records Into Evidence
Agency records and published indicators prepared for analysis: sources linked, definitions checked across the years, denominators fixed, and the policy question answered with estimates and intervals traceable to the records.
No Commitment Required
Strictly Confidential
In-House PhD-Level Statisticians
Fixed Project Fee
Same-Day Response
What Is Involved in Analyzing Policy Records?
Government administrative data analysis starts from records that were created to run a program, not to answer a research question: enrollment files, case records, benefit and tax data, and indicators published by jurisdiction. The work therefore begins with the data themselves, with what each row represents, how sources were linked, how definitions changed over the years and who is missing, before any policy effect is estimated. OnlineSPSS, a specialist statistical consultancy for academic, clinical and professional research, has in-house PhD-level statisticians build the analysis file and log how every outcome was constructed. The policy conclusions drawn from the estimates are the researcher’s to defend.
WHEN TO ENGAGE
Agency Data Before and After Access
Policy projects often need a statistician earlier than researchers expect, because the way the data were constructed decides what can be estimated.
Records Obtained or Requested An agency has granted or is about to grant access, and the record layout, the variables on offer and the linkage keys decide which questions the data can address.
Sources to Be Linked Records from two or more agencies must be matched, and the match rate, the unmatched cases and the risk of false links have to be established before analysis.
Outcome Definition Unsettled The outcome could be defined as an event, a rate or a duration, and the definition must be fixed before it is related to the policy of interest.
Findings Under Scrutiny A report or paper rests on agency data, and a reviewer or the agency has asked how records were selected, counted and treated when missing or revised.
RECORD-BASED ANALYSIS
From Agency Records to a Policy Estimate
The sequence moves from what a row means to what a policy changed, and the estimate is only as clear as the first stage.
Unit of Record
Each file is described by what one row stands for, whether a person, a case, an event or a place-month, and files are reshaped to the unit the question needs.
Linkage Between Sources
Records matched across agencies are linked by stated rules, and the share linked, the share unmatched and the characteristics of each group are reported.
Outcomes and Denominators
Policy outcome data are defined as counts, rates or durations with a fixed denominator, and a change in coding rules is treated as a break in the series.
Places With and Without the Rule
Where the records cover places with and without a new rule, Difference-in-Differences (DiD) compares their outcomes before and after it, with each place's adoption date verified from the records.
One Start Date
For a single jurisdiction with a known start date, the earlier trend is extended and compared through Interrupted Time Series, with level and slope changes given intervals.
Policy Reporting
Reporting draws on the conventions of Public Policy & Administration, one of the fields gathered under Education & Social Sciences.
COMMON ISSUES
Faults in Administrative Data That Alter Findings
These faults live in the records, and examining the data before the policy variable is introduced brings them out.
A Definition Changed Mid-Series A diagnosis code, eligibility rule or reporting form was revised partway through the years, and the jump this causes in counts can be mistaken for the effect of a policy.
Counts Without Denominators Event counts are compared across places of different size, or population figures come from another year, so the comparison reflects population and not the policy.
Late-Arriving Records The latest months look low because records are still being filed, and a fall at the end of the series is read as an effect of the policy.
Persons and Events Confused A person with several records is counted several times, so a rate per record is reported as though it were a rate per person.
Small Cells Suppressed Counts below a disclosure threshold are masked or rounded, and treating masked cells as zero or dropping them can bias the totals for small places.
Unlinked Records Ignored People who fail to match across sources may differ from those who do, so analyzing matched records alone can describe a selected group without saying so.
DELIVERABLES
Documents That Trace an Estimate to a Record
Decisions about rows, matches and definitions are recorded alongside the estimates, so the path from extract to report stays visible after delivery.
Record Layout Summary For every file, what a row stands for, the key variables, the period covered and the number of rows at each stage of cleaning.
Linkage Report The matching rules, the share of records linked and unmatched, and a comparison of the two groups on the characteristics available.
Outcome Construction Note How each outcome and denominator was built, with any change in coding or reporting rules marked on the timeline.
Policy Effect Estimates Estimates in the units the outcome is recorded in, with 95% confidence intervals, the comparison units used and the periods covered.
Data Handling Statement A statement of exclusions, suppressed cells and late-arriving records, prepared for a methods section or an appendix.
Processing Code for New Releases A rerunnable chain from each source extract to each table, so a revised release from the agency can be processed without redoing the decisions.
ILLUSTRATIVE EXAMPLE
A State Rule Change Seen Through Enrollment Records
An illustration of how a change in reporting practice is handled in an evaluation of a school meal rule.
ILLUSTRATIVE EXAMPLE · SCHOOL MEAL ELIGIBILITY POLICY EVALUATION · UNITED STATES
A doctoral researcher in public policy wanted to know whether a state rule widening eligibility for free school meals was followed by higher meal participation, using monthly district records from several states.
The unit of record and the outcome were defined first, as participating children per enrolled child in each district and month. The complication was that districts in one state changed how they reported participation partway through the period, which produced a jump unrelated to the rule. That date was marked as a break, the outcome was rebuilt from a field available in both periods, and the latest months were left out because records were still arriving. Districts in states without the rule served as the comparison, and the result was reported as a change in participation with its confidence interval, worded as an association given the design.
Outcome: The researcher received the record layout summary, the outcome construction note, the estimated change in participation with its 95% confidence interval, the data handling statement and the code.
This example does not describe a real client or project. We never share details of actual client work, in line with the confidentiality we promise every researcher. This illustrative example is included to show the type, scope, and standard of work involved in this kind of analysis.
SOFTWARE
Handling Large Record Files in Code
The program is chosen for the size of the extracts and for the code your agency partner or department can read, and the work is scripted from the source files.
HOW WE ASSESS YOUR PROJECT
We Review Your Actual Research
Your assessment is based on a review of your actual research materials and analytical requirements, not a generic project category.
You Submit Your Project Details
Share whatever is available at the current stage, such as the research aim, protocol or proposal, dataset, variable list, manuscript or statistical requirements.
We Review Your Research and Define the Scope
Your statistically trained Research Coordinator reviews the materials to understand the study design, research questions, data structure, analytical requirements and expected deliverables. If clarification is needed, additional documents or information can be requested before the assessment is finalized.
Your Analysis Is Completed and Delivered
Once the scope and fixed project fee are agreed, the in-house statistician completes the work and the Research Coordinator remains the communication point through delivery, with the analysis, deliverables and timeline as agreed.
RESEARCH RESPONSIBILITY
What Stays With the Policy Researcher
Permission to use agency records, the policy question and any recommendation rest with the researcher, and record preparation and estimation are agreed in the scope.
Agreed Statistical Support OnlineSPSS provides the agreed methodological, analytical and statistical reporting support, defined in a scope that is confirmed before any work begins.
No Intellectual Contribution OnlineSPSS does not claim any intellectual contribution to your research. Authorship of the report, article or thesis remains yours.
Your Research Decisions The policy question, the permissions to use the records, the interpretation of the results and the recommendations remain yours, with overall responsibility.
Confidential Handling Research data and project details are handled confidentially, project information is not published, and an NDA is available where required.
FREQUENTLY ASKED QUESTIONS
Common Questions About Policy Records
Agency records were kept for running programs, so the questions that matter most concern how those records were coded, matched and protected.
What does statistical support for administrative data in a policy study cover?
Government administrative data analysis covers the construction of the analysis file from agency records, the definition of outcomes and denominators, the estimation of policy effects and the documentation of each decision. The estimates are reported with intervals, and the data problems that could change them are stated alongside.
How should a change in how a variable was recorded be handled?
The date of the change is established first and treated as a possible break in the series. Where a field exists in both periods, the outcome is rebuilt from it so that the definition is constant. Where none exists, the periods are analyzed separately or the break is modeled explicitly, and a comparison unit that was not affected shows whether a shift came from the recording change.
How are records linked across agencies and checked?
Records are matched on a shared identifier where one exists, or on combinations of fields by stated rules where it does not. The report gives the share linked and unmatched, compares the two groups on available characteristics, and shows how the main estimates change if the matching rule is made stricter or looser.
Are suppressed small counts a problem for analysis?
They can be. Masked or rounded cells are neither zero nor missing at random, and dropping them can bias results for small places. Options include combining places into larger units, treating a masked count as known to lie within a range, and reporting results with and without the affected places, always within the disclosure rules that apply to the data.
How do you protect my research data and confidentiality?
Your data and project details are handled confidentially, and project information is not published. Researcher names and institutions are not used publicly without permission. An NDA is available where required and can be arranged before sensitive materials are shared. Read more about confidentiality and data security.
What happens when I request an assessment, and how long does it take?
You share your research question, design and relevant materials, such as the record layout, the terms under which the data may be used and an extract or sample, and a statistically trained Research Coordinator reviews the request and responds the same day. The analysis, deliverables, timeline and one fixed project fee are then confirmed before any work begins, so the timeline follows from the agreed scope.
Still have a question? Find more answers in our full FAQ.
NEXT STEP
Records Kept to Run
Programs, Read for Research
Agency files answer the questions they were kept for, and turning them toward a policy question starts with knowing how each file was built and coded.
No Commitment Required
Strictly Confidential
Fixed Project Fee
In-House PhD-Level Statisticians
Same-Day Response
Serving Researchers Since 2014