RESEARCH SUPPORT
Categorical data analysis support
when one chi-square test is not enough
Categorical outcomes are analyzed with stratified tables when sites or subgroups differ, trend tests when categories are ordered, exact methods for sparse cells, and logistic or log-linear models when one table cannot hold every variable, each with an interval.
No Commitment Required
Strictly Confidential
In-House PhD-Level Statisticians
Fixed Project Fee
Same-Day Response
What Does Categorical Data Analysis Involve?
Categorical data analysis support is for the point where one two-way table no longer holds the question: a third variable, ordered categories, sparse cells or records that are not independent. A Cochran-Mantel-Haenszel test combines stratified tables, and log-linear models for contingency tables describe how several categorical variables relate when no single outcome is designated. Exact or ordinal methods take over where counts are small or the categories are ordered. As a specialist statistical consultancy for academic, clinical and professional research, OnlineSPSS places the choice of method with in-house PhD-level statisticians, and each effect is reported with its interval, with the reason for every choice of method recorded beside it.
WHEN TO ENGAGE
Counts Whose Analysis Is Not Settled
The structure of the table, the pairing of the observations and the number of variables each point to a different analysis, so the first task is to find out which of them applies.
Small or Sparse Tables Some cells hold very small counts, and the researcher must choose between an exact approach, merged categories and a model, with the reasoning stated and fixed in advance.
Matched Sets or Repeated Visits Cases were matched to several controls, or the same people were classified at several visits, and the analysis must keep that structure instead of treating every row as an independent observation.
Other Variables to Hold Constant The comparison between categories may be confounded, so the analysis must move from a two-way table to a model with covariates and report adjusted effects.
Ordered or Many Categories The outcome has ordered levels or many categories, and splitting it at one cutpoint or treating it as unordered would waste information or hide a trend.
TABLE AND MODEL CHOICES
Pairing, Cell Counts and Covariates Decide the Method
A single test of independence suits separate observations, adequate counts and a question with no other variable to adjust for, and most of the methods below serve data that fall outside one of those conditions.
Where the Basic Tests End
The standard tests, Chi-Square Test, Fisher's Exact Test and McNemar's Test, come first, and the cards below begin where they stop.
Stratified Tables
A Cochran-Mantel-Haenszel test combines 2 by 2 tables across strata such as sites into one common odds ratio, after a check that the odds ratio is similar across the strata.
Matched Sets and Repeated Visits
Cases matched to several controls are analyzed with conditional logistic regression, and repeated visits per person with Generalized Estimating Equations (GEE), so neither is reduced to one table.
Adjusted Comparisons
Logistic Regression estimates odds ratios with other variables held constant, and its assumptions and sparse-data problems are checked before the model is accepted.
Log-Linear Structure
Where no single outcome is designated, log-linear models describe how several categorical variables are associated, with terms added and tested in an order stated beforehand.
Ordered Outcomes
Ordered categories are analyzed with a trend test or a model for ordered outcomes, so that the order contributes information and a single cutpoint is not forced on the data.
TABLE PITFALLS
Habits That Misstate Categorical Results
A table of counts looks simple, which is why misreadings of its percentages, pairings and strata so often go unchallenged.
Percentages on the Wrong Margin Row percentages are shown where column percentages answer the question, or the reverse, so the same table appears to tell two different stories to a reader.
Matched Sets Collapsed Each case has several matched controls, or the same people sit in several matched sets, and the data are collapsed into one table, so the matching is lost and the odds ratio can be distorted.
Zero Cells in a Logistic Model A category with no events drives a logistic estimate toward infinity, and the software may still print a coefficient, so sparse predictors are checked before any odds ratio is read.
Association That Reverses A table pooled across sites or subgroups shows an association that reverses within each of them, so the strata are examined before any combined table is reported.
Cutpoints Chosen After Looking A continuous or ordered measure is split at the value that gives the clearest contrast, which inflates the evidence and rarely reproduces in a new sample.
Visits Counted as Persons Counts from repeated visits of the same individuals are cross-tabulated as if each visit were a separate person, so the sample size is overstated and the interval too narrow.
DELIVERABLES
A Method Plan, Annotated Tables and Effect Measures
The documents show which question each table answers and how every reported proportion, difference or ratio was obtained.
Method Selection Note The design features that decided the method, namely pairing, counts per cell, ordering and covariates, with the reason that each alternative was set aside.
Annotated Contingency Tables Counts with the percentages based on the correct margin, the denominator labeled in each table and the minimum expected count reported where it matters.
Effect Measure Summary Differences in proportions, risk ratios or odds ratios as the design allows, each with a 95% confidence interval and a note on which direction it describes.
Model Specification Record For logistic and log-linear models, the terms entered, the reference categories, the checks for sparse data and the fit of the model to the observed counts.
Sparse-Cell Report Any cell with few observations, the way it was handled and a comparison showing whether the conclusion depends on that handling, so that a reviewer can judge the sensitivity.
Reporting Wording for Results Statistical wording for each comparison with its estimate and interval, matched to the output and the style your field expects, for you to check and incorporate.
ILLUSTRATIVE EXAMPLE
Infection Compared Across Four Hospital Sites
An illustration of how a table pooled across sites is replaced by a stratified analysis that keeps each site's comparison intact.
ILLUSTRATIVE EXAMPLE · SURGICAL SITE INFECTION RESEARCH · UNITED STATES
A nurse researcher compared surgical site infection between two wound dressing protocols, using records from four hospital sites that had each adopted the protocols to a different extent.
Pooled into one table, the newer dressing looked worse, but the sites that used it most also operated on the highest-risk patients. Site was therefore treated as a stratum: a Cochran-Mantel-Haenszel analysis gave a common odds ratio after a check that the odds ratios were similar across sites. The complication was that one small site recorded no infections under one protocol, so the stratified estimate was compared with an exact analysis and with that site set aside. Wound severity, recorded in three ordered grades, was analyzed with a trend test, not split at one grade, and the finding was worded as an association within sites, not as an effect of the dressing.
Outcome: The researcher received the method selection note, the tables by site, the common odds ratio with its 95% confidence interval, the sparse-cell comparison and the reporting wording for the results.
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
Exact, Paired and Model-Based Analyses of Counts
Tables of counts and models for categories run in each of these programs, though exact tests and log-linear fitting sit in different procedures.
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
Category Definitions Come From the Research Team
How an outcome was defined and what a difference in proportions means in practice are questions for the research team, while the analysis reports what the data show.
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 manuscript or thesis remains yours.
Your Research Decisions The definition of each category, the clinical or practical meaning of a difference, the discussion and the conclusions 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
Categorical Outcomes: Tables, Models and Percentages
How a table of counts should be read depends on whether its levels are ordered, how many variables are involved and which margin the percentages use.
What does categorical data analysis support cover?
Categorical data analysis support covers the choice of method for outcomes in categories, including exact tests, paired comparisons, logistic and log-linear models, and the reporting of effect measures with confidence intervals. The design decides the method, namely independent or paired observations, the size of the counts, the number of variables and whether the categories are ordered.
When is a log-linear model better than a series of separate tests?
When there is no single outcome and the interest lies in how several categorical variables are associated. Separate tests on pairs of variables can mislead, since an association between two may arise through a third. A log-linear model examines the variables together and tests which associations are needed, with the terms entered in an order chosen beforehand.
Can an ordered categorical outcome be analyzed with an ordinary table test?
It can, but the test ignores the order and so wastes information. A trend test or a model for ordered outcomes uses the order and is usually more sensitive to a gradual change across levels. The choice is made from the research question, and collapsing to a yes or no outcome is reserved for questions that are themselves a threshold.
When is a Cochran-Mantel-Haenszel test better than one pooled table?
A Cochran-Mantel-Haenszel test is preferred when the same comparison is made in several strata, such as sites, and a pooled table could mix differences between strata with the comparison itself. It gives one common odds ratio across strata, which is reported only when the stratum odds ratios are similar; where they differ, the estimates are reported by stratum.
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 outcome definitions, the cross-tabulations you hold and the dataset, 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
Categorical Outcomes Analyzed
as the Design Requires
Whether a comparison holds within each site or stratum, and how many observations fall in the smallest cell, are the two facts that most often change which analysis is appropriate.
No Commitment Required
Strictly Confidential
Fixed Project Fee
In-House PhD-Level Statisticians
Same-Day Response
Serving Researchers Since 2014