Executive Research Companion
Chi-Square Intelligence for Healthcare Operations
From categorical counts to accountable decisions: a leadership framework for detecting, diagnosing, and acting on operational variation in access, quality, imaging, revenue cycle, and workforce data.
Leadership Conclusion
The chi-square test is most valuable when a healthcare leader must determine whether an observed distribution of categorical events differs from a plan, whether two categorical variables are associated, or whether operational patterns vary across sites. Its value is not just the p-value. Decision quality depends on validated categories, independent observations, adequate expected counts, effect size, residual analysis, and a clinically or operationally meaningful response.
Healthcare operations generate categorical data continuously: attended or missed appointments, completed or incomplete examinations, accepted or denied claims, event classifications, discharge destinations, compliance status, and reason codes. Chi-square methods convert these counts into a formal test of whether observed differences are compatible with chance variation under a defined null model. The test answers a narrow but consequential question. It does not identify causation, adjust for confounding, measure economic value, or prove clinical importance. A statistically significant result should open an operational investigation rather than close one.
The three-part executive reading rule
Detect
Use the omnibus chi-square statistic to determine whether the full categorical pattern is inconsistent with the null model.
Diagnose
Compare observed and expected counts, then inspect adjusted residuals to identify the categories and cells driving the result.
Decide
Pair effect size and absolute differences with patient impact, cost, feasibility, equity, and balancing measures before authorizing change.
Six conclusions for healthcare leaders
Start with the decision, not the software
Define the operational question, target population, categories, denominator, comparison basis, and action threshold before calculating a statistic.
Select the correct chi-square design
Goodness-of-fit compares an observed distribution with a specified expectation. Independence tests association between two categorical variables in one population. Homogeneity compares a distribution across independent groups.
Treat data structure as a governance issue
Duplicate encounters, repeated measures, unstable reason codes, and missing categories can invalidate the analysis even when the computation is technically correct.
Report more than statistical significance
A decision-ready result includes counts, percentages, expected counts, chi-square, degrees of freedom, p-value, effect size, residuals, and operational interpretation.
Escalate when the question is more complex
Use Fisher’s exact test for sparse, small tables; McNemar’s test for paired binary outcomes; regression for adjusted analysis; and longitudinal or clustered methods when independence is violated.
Link findings to intervention and re-measurement
The analysis should specify the process owner, targeted cells, intervention, balancing measures, and post-change monitoring period.
Where the leadership questions live
| Healthcare leadership question | Primary test | Example categorical outcome | Decision use |
|---|---|---|---|
| Does demand match the staffing plan? | Goodness of fit | CT volume by shift | Rebalance coverage or scheduling templates |
| Is access associated with scheduling lead time? | Independence | Completed vs missed appointment | Target reminders, navigation, or slot release |
| Do denial patterns differ by hospital? | Homogeneity | Denial reason category | Assign site-specific revenue-cycle countermeasures |
| Did an intervention change a paired binary outcome? | Not ordinary chi-square | Same unit before vs after | Use McNemar or a repeated-measures model |
Executive Lens
Chi-square is a detector and localizer of categorical variation, not a causal model. A credible analysis should identify what differs, where it differs, how much it differs, and what operational hypothesis should be tested next.
The statistical foundation
Observed counts are the actual numbers in the contingency table. Expected counts are generated from the null hypothesis. Squaring prevents positive and negative differences from canceling, while dividing by the expected count scales the discrepancy. The sum is compared with a chi-square reference distribution whose shape depends on the degrees of freedom.
The core statistic
χ² = Σ (O − E)² / E
Each cell contributes a nonnegative amount. Larger discrepancies, especially where the expected count is small, contribute more strongly to the total.
Expected counts
Eij = (row total × column total) / N
For independence and homogeneity, expected counts represent the table anticipated if row and column classifications were unrelated.
Degrees of freedom
df = k − 1 · df = (r − 1)(c − 1)
Goodness-of-fit uses k fixed categories. An r by c contingency table uses cell information remaining after marginal constraints.
What the p-value actually is
The shaded area represents results at least as discrepant as the critical value. The p-value is not the probability that the null hypothesis is true.
A small p-value indicates that the observed table would be unusual if the null model and test assumptions were correct. It does not measure effect magnitude, decision value, or the probability of replication. The American Statistical Association and subsequent methodological guidance caution against using a single threshold as a binary decision rule (Greenland et al., 2016; Wasserstein & Lazar, 2016).
The inferential boundary
The test can support
- Evidence that an observed distribution differs from a specified expectation
- Evidence of association between categorical variables
- Evidence that group distributions are not homogeneous
- A global comparison across all table cells
Not established without additional analysis
- The clinical, financial, or operational importance of the difference
- Causation, or the mechanism producing the association
- Which cell differences are responsible, unless residuals or follow-up comparisons are examined
- Adjustment for age, acuity, payer mix, site clustering, repeated encounters, or other confounders
Methodological Position
The report treats the chi-square test as an inferential component within an operational decision-making system. It rejects three common shortcuts: equating p < .05 with importance, inferring causation from association, and treating a large sample as a substitute for sound category design.
Interactive design selector
Selection begins with the sampling design and the operational question. Independence and homogeneity use the same general computation, but their interpretations differ. Choose the statement that matches your situation.
Goodness of fit df = k − 1
Use goodness of fit when one set of observed category counts is compared with a distribution specified in advance: observed demand by shift versus a staffing model, actual appointment mix versus a capacity plan, or event-type counts versus an established benchmark. The expected distribution must not be reverse-engineered after observing the data. Effect size: Cohen’s w.
Worked example: Case 1, CT demand versus the staffing plan.
Test of independence df = (r − 1)(c − 1)
Use independence when each observation from one population is classified into two categorical variables: appointment status by lead-time band, examination completion by modality, or safety-event severity by shift. A significant result indicates association, not direction or causation. When one variable is ordered, supplement the omnibus test with a trend test or regression. Effect size: Cramer’s V or phi.
Worked example: Case 2, imaging nonattendance by scheduling lead time.
Test of homogeneity df = (r − 1)(c − 1)
Use homogeneity when comparing independent groups on the distribution of a single categorical outcome: denial reasons across hospitals, discharge destinations across service lines, or patient-experience response categories across access channels. A residual analysis should follow the global test to identify site-category combinations that depart from expectations. Effect size: Cramer’s V.
Worked example: Case 3, denial-reason distributions across three hospitals.
Escalate: not ordinary chi-square
Binary outcomes measured on the same units before and after are paired, not independent. Use McNemar’s test, or conditional and mixed modeling for more complex repeated designs. Applying an ordinary chi-square here understates uncertainty and can produce a p-value that is too small.
See the escalation table on the Interpretation tab for the full alternatives map.
From the dashboard difference to the inferential question
| Dashboard observation | Null hypothesis | Chi-square question | Leadership response if evidence is credible |
|---|---|---|---|
| Evening CT volume appears above plan | Demand follows the planned shift distribution | Goodness of fit | Validate acuity and arrival drivers, then adjust staffing or protocols |
| Long-lead appointments appear more likely to be missed | Attendance is independent of lead-time category | Independence | Target long-lead bookings with reminders, waitlist fills, or navigation |
| Authorization denials appear concentrated at one hospital | Denial-reason distributions are equal across sites | Homogeneity | Investigate site-specific registration, authorization, and documentation workflows |
Common Design Error
The ordinary chi-square test assumes independent observations. If the same patient, clinician, unit, or hospital contributes repeated observations, the nominal p-value may be too small because the effective information is less than the row count suggests.
Pre-analysis data readiness
Statistical correctness begins before the contingency table is built. Operational databases are encounter-centered, longitudinal, and frequently revised. A leader should require a brief analytical specification covering unit of analysis, population, period, categories, denominator, exclusion rules, and the treatment of missing or repeated observations. Work through each control below and confirm it has an explicit answer.
A calculation should not proceed to executive interpretation until every control has an explicit answer.
Unit of analysis
Patient, appointment, claim, examination, event, or response?
Failure mode: mixing units creates invalid totals. Safeguard: state one unit in the analytic charter.
Mutual exclusivity
Can one observation occupy more than one category?
Failure mode: double-counting inflates sample size. Safeguard: use a hierarchy or one primary classification.
Exhaustiveness
Where do unknown, pending, and other observations go?
Failure mode: selective omission changes proportions. Safeguard: report missing and unclassified counts.
Independence
Can the same patient or site contribute repeated observations?
Failure mode: standard errors are understated. Safeguard: deduplicate, aggregate, or use clustered methods.
Expected counts
Are cells sufficiently populated for the approximation?
Failure mode: p-values can be inaccurate. Safeguard: use exact or simulation-based alternatives.
Stable definitions
Did codes or workflows change during the period?
Failure mode: apparent differences reflect taxonomy drift. Safeguard: version categories and document change dates.
Denominator
Are percentages based on comparable opportunities?
Failure mode: counts obscure exposure differences. Safeguard: report both counts and relevant rates.
Expected-count guidance
The familiar statement that every cell must contain at least five observations is an oversimplification. The issue concerns expected counts and the adequacy of the asymptotic approximation. A common practical rule: no expected count below 1, and no more than approximately 20% below 5. For sparse 2-by-2 tables, Fisher’s exact test or other exact procedures should be considered; simulation evidence also shows that default continuity corrections are not universally optimal (Campbell, 2007; Kim, 2017; Lydersen et al., 2009).
Missing data
Missing categories should be quantified rather than silently dropped. Report unknown and unclassified counts alongside the analytic table so proportions remain interpretable.
Category collapse
Categories may be combined only when the new grouping is substantively coherent and prespecified. Collapsing categories to obtain a desired p-value changes the question and compromises interpretability.
Goodness of fit Worked healthcare case 1 · Illustrative synthetic data
Does CT demand match the staffing plan?
A hospital planned CT staffing around a 40% day, 40% evening, and 20% night demand distribution. During an illustrative month, 600 examinations were completed. Shift is compared with a distribution specified before the observations were analyzed, so this is a goodness-of-fit question.
Observed CT volume compared with the staffing plan distribution. The overall difference is statistically detectable, but the pattern is operationally concentrated in a 30-examination shift from day to evening.
| Shift | Observed | Expected | O − E | Cell contribution |
|---|---|---|---|---|
| Day | 210 | 240 | −30 | 3.75 |
| Evening | 270 | 240 | +30 | 3.75 |
| Night | 120 | 120 | 0 | 0.00 |
| Total | 600 | 600 | 0 | 7.50 |
The statistical evidence challenges exact alignment with the staffing distribution, but the magnitude is small. The operationally relevant fact is that evening demand exceeded plan by 30 examinations while day demand was 30 below plan. A leader should examine complexity-weighted workload, turnaround time, overtime, and arrival sources before moving staff. Counts alone do not establish the number or skill mix of resources required.
Decision Translation
Do not respond to p = .024 by automatically adding evening staff. First determine whether the 30-examination variance is persistent, whether it affects turnaround or safety, and whether demand can be shifted through scheduling design. Then test a staffing or template intervention with balancing measures.
Test of independence Worked healthcare case 2 · Illustrative synthetic data
Is imaging nonattendance associated with scheduling lead time?
Missed imaging appointments delay diagnosis, reduce asset utilization, and waste specialized labor. Among 720 scheduled outpatient imaging appointments, is completion status associated with the time between scheduling and the appointment? Under the null hypothesis of independence, every lead-time row would hold 160 completed and 20 missed appointments because all four row totals are equal.
| Lead time | Completed | Missed | Total | Missed rate |
|---|---|---|---|---|
| 0-3 days | 174 | 6 | 180 | 3.3% |
| 4-7 days | 169 | 11 | 180 | 6.1% |
| 8-14 days | 158 | 22 | 180 | 12.2% |
| 15+ days | 139 | 41 | 180 | 22.8% |
| Total | 640 | 80 | 720 | 11.1% |
Missed appointment rates rise monotonically from 3.3% to 22.8% across lead-time categories.
The 15+ day category accounts for 24.81 of the total 40.61 statistic, approximately 61%. A global p-value alone would conceal this operational localization.
Which cells drive the result?
Adjusted standardized residuals localize the association. Values approximate z-scores under the null model: positive means more observations than expected, negative means fewer.
Leadership interpretation
The long-lead group shows approximately 6.8 times the observed missed-appointment rate of the 0-3 day group. That ratio is descriptive and unadjusted. The next analysis should use logistic regression or a related model to account for modality, prior missed visits, age, insurance, socioeconomic context, referral source, day of week, and repeated appointments. Operational action may include waitlist automation, confirmation escalation, transportation screening, patient navigation, or reserved short-lead capacity. Interventions should be designed to avoid labeling patients as noncompliant when the underlying constraint is structural access.
Executive Reporting Language
Appointment completion was associated with scheduling lead time, χ²(3, N = 720) = 40.61, p < .001, Cramer’s V = .24. Adjusted residuals localized excess missed appointments to the 15+ day group. Because the analysis is unadjusted and encounter-based, the findings should guide targeted investigation and prospective testing rather than causal attribution.
Test of homogeneity Worked healthcare case 3 · Illustrative synthetic data
Are denial-reason distributions homogeneous across hospitals?
Three hospitals each recorded 200 denied claims. Is the distribution of denial reasons the same across sites, or do particular hospitals show distinctive patterns requiring site-specific intervention? Aggregate denial rates can conceal distinct causal structures across sites.
Observed denial reasons across three hospital sites, 200 denied claims per site. With equal site totals, each row expects 41.67 eligibility, 50 authorization, 43.33 medical-necessity, 45 coding and documentation, and 20 other denials.
Adjusted residuals reveal where distributions diverge
Residual magnitudes near or above 2 identify cells that differ materially from the null model. Positive values indicate more denials than expected, negative values fewer.
North shows substantially more authorization denials than expected and fewer eligibility denials. South shows the opposite pattern. Central displays no equally dominant deviation. A systemwide intervention aimed at only one denial reason would be inefficient.
Translating residuals into targeted revenue-cycle action
| Site signal | Operational hypothesis | Validation | Potential intervention |
|---|---|---|---|
| North: authorization excess | Scheduling or authorization workflow is incomplete or late | Audit lead time, payer rules, order completeness, and handoffs | Authorization work queue, escalation standard, pre-service checklist |
| South: eligibility excess | Coverage verification or registration data are unreliable | Audit eligibility response, demographics, plan changes, and registration rework | Real-time eligibility, registrar feedback, hard stops for missing fields |
| Central: diffuse pattern | No single reason dominates | Review severity, payer mix, appeal yield, and low-frequency subcategories | Maintain broad controls and monitor for emerging concentration |
Leadership Conclusion
The test supports differentiated intervention. North requires an authorization-centered investigation, South requires an eligibility-centered investigation, and Central requires broader monitoring. The effect size indicates that site is informative but not determinative; payer and service-line stratification remain important.
Live chi-square calculator
Run the full decision-ready computation on your own counts: statistic, degrees of freedom, exact p-value, effect size, expected counts, adjusted residuals, and an expected-count adequacy screen. The engine reproduces every result in the three worked cases. Illustrative use only; validate against local data and governance review before operational decisions.
Compare observed counts with expected percentages specified in advance. Preloaded with Case 1, CT demand by shift.
Test independence or homogeneity on an r by c table of observed counts. Preloaded with Case 2, appointment status by lead time.
Validation Control
Reproduce at least one small table by hand or in a second validated tool. Confirm the same N, expected counts, chi-square, degrees of freedom, p-value, and effect size before presenting the result.
Interpretation beyond the p-value
Healthcare leaders should reject the false choice between statistical rigor and practical relevance. Both are required. A p-value assesses compatibility with the null model, while effect size and operational consequences address magnitude and decision value. Large datasets can make small discrepancies statistically detectable; small datasets can conceal operationally important differences through low power.
Cramer’s V
V = √[χ² / (N × min(r − 1, c − 1))]
Ranges 0 to 1 and standardizes association strength for contingency tables. For 2-by-2 tables, phi is common; for goodness-of-fit, use Cohen’s w.
Pearson residual
r = (O − E) / √E
Adjusted standardized residuals also account for marginal proportions and are preferable for localizing unusual cells.
Cohen’s w
w = √[Σ(pobs − pexp)² / pexp]
Goodness-of-fit effect size computed on proportions rather than counts.
Four layers of a decision-ready interpretation
Statistical
Is the observed table unusual under the null model? Requires chi-square, df, p-value, and assumption checks.
Magnitude
How strong is the discrepancy or association? Requires Cramer’s V, phi, or Cohen’s w plus absolute percentage differences.
Diagnostic
Which cells or categories drive the result? Requires observed and expected counts, adjusted residuals, and planned follow-up.
Operational
What should change, who owns it, and what could be harmed? Requires process knowledge, cost, patient impact, feasibility, and balancing measures.
Sample size and power planning
Approximate N required for df = 3, alpha = .05, and 80% power. Required N grows rapidly as the target effect becomes smaller; real studies require category-specific planning.
A nonsignificant result may mean either that the null model is reasonably compatible with the data or that the study lacks power to detect the effect that matters. Sample-size planning should begin with the smallest operationally important effect, the intended degrees of freedom, alpha, power, expected category proportions, and the feasibility of adequate cell counts (Rahman et al., 2025). Unequal marginal distributions, sparse categories, multiple comparisons, clustering, and attrition all increase the required sample. Leaders should also avoid collecting an enormous sample merely to declare a trivial difference significant.
Planning Rule
Define the minimum difference that would change an operational decision. Power the study to detect that difference, then confirm that each planned category will have sufficient expected observations. If the minimum actionable difference is not defined, the sample-size calculation is not decision anchored.
When the ordinary chi-square is not the best primary method
| Data situation | Why ordinary chi-square is insufficient | Preferred direction |
|---|---|---|
| Sparse 2-by-2 table | Asymptotic p-value may be inaccurate | Fisher’s exact or another exact or unconditional test |
| Same patients before and after | Binary outcomes are paired, not independent | McNemar’s test; conditional or mixed modeling |
| More than two repeated binary measurements | Within-person correlation | Cochran’s Q, GEE, or mixed-effects model |
| Ordered exposure or outcome | Omnibus chi-square ignores monotonic trend | Cochran-Armitage trend test or ordinal and logistic regression |
| Need to adjust for case mix | Bivariate association may be confounded | Binary, multinomial, or ordinal logistic regression |
| Counts with unequal exposure time | Raw counts are not comparable to rates | Poisson or negative-binomial regression with offset |
| Patients nested in clinicians or hospitals | Cluster correlation understates uncertainty | Cluster-robust, GEE, or multilevel model |
| Monthly performance monitoring | Repeated significance tests ignore temporal structure | Statistical process control, interrupted time series, or longitudinal model |
For 2-by-2 tables, methodological research shows that test choice is more nuanced than the simplistic rule of using Fisher when n is small. The sampling scheme, desired error properties, expected counts, and availability of exact unconditional or mid-p procedures all matter (Campbell, 2007; Lydersen et al., 2009). Consultation with a statistician is appropriate when events are rare, tables are sparse, or the analysis will inform high-stakes clinical or financial policy.
The healthcare leader’s operating cycle
A chi-square analysis should be embedded in an operating cycle that connects measurement to intervention. The cycle begins with the decision and ends with re-measurement, because a one-time significance test does not establish sustained improvement.
1 · Frame
State the decision, unit of analysis, population, categories, expected pattern, and action threshold.
2 · Validate
Confirm source systems, duplicate handling, missingness, category stability, independence, denominators, and expected counts.
3 · Test
Calculate the appropriate statistic, degrees of freedom, p-value, and effect size with reproducible code or validated software.
4 · Diagnose
Compare observed and expected counts; inspect adjusted residuals; apply planned post hoc controls; stratify where process knowledge warrants.
5 · Act
Translate the localized signal into a workflow hypothesis, owner, intervention, implementation date, and balancing measures.
6 · Re-measure
Use an appropriate post-change design. For ongoing monthly monitoring, supplement testing with statistical process control or longitudinal methods.
Leadership review questions
Check off each question your team can answer before authorizing action on a chi-square result.
0 of 9 answered
Minimum reporting elements
| Element | Minimum content | Why leaders need it |
|---|---|---|
| Question and design | Population, period, unit, variables, and selected chi-square type | Confirms that the method matches the decision |
| Observed data | Counts plus row or column percentages | Shows the actual operational pattern and denominator |
| Expected data | Expected counts and adequacy check | Demonstrates that the approximation is defensible |
| Test result | Chi-square, df, N, exact p-value or p < .001 | Quantifies evidence against the null model |
| Magnitude | Cramer’s V, phi, or Cohen’s w plus absolute differences | Separates detectable from important |
| Localization | Adjusted residuals and controlled post hoc comparisons | Identifies actionable categories |
| Limitations | Missingness, repeats, confounding, sparse cells, multiple testing, generalizability | Prevents overclaiming |
| Action | Owner, intervention, timeline, outcome, and balancing measures | Converts analysis into accountable improvement |
Governance, equity, and ethical interpretation
Categorical analyses can identify disparities, but they can also reproduce biased classification, missingness, and structural differences in access. A significant association between a patient characteristic and nonattendance should not be interpreted as an individual deficit. It may reflect transportation, work schedules, digital access, language, caregiving obligations, referral design, or the location and timing of services (Cleary et al., 2024; Mander et al., 2018).
Governance safeguards
- Use the minimum necessary data and de-identify operational extracts whenever feasible
- Review small cells for privacy risk as well as statistical instability
- Document category definitions, mappings, and changes in the data dictionary
- Require independent analytic review when results will influence high-stakes clinical, employment, or financial policy
Equity practices
- Avoid using protected characteristics as simplistic causal explanations or punitive targeting variables
- Include patient and frontline perspectives when translating an association into an intervention
- Measure whether the intervention changes access, experience, or burden differently across groups
Equity Principle
The operational goal is to redesign barriers, not to classify patients as the problem. Chi-square can reveal where categorical outcomes differ, but leadership must investigate the system that produces those differences.
Final Leadership Takeaway
Use chi-square to detect and localize categorical operational variation. Use process knowledge, effect size, adjusted analysis, and patient-centered judgment to decide what that variation means and what should change.
References
APA 7th edition. DOI links are provided for verification and access. Filter by domain.
Campbell, I. (2007). Chi-squared and Fisher-Irwin tests of two-by-two tables with small sample recommendations. Statistics in Medicine, 26(19), 3661-3675.
doi.org/10.1002/sim.2832Chandawarkar, R., Nadkarni, P., Barmash, E., Thomas, S., Capek, A., Casey, K., & Carradero, F. (2024). Revenue cycle management: The art and the science. Plastic and Reconstructive Surgery – Global Open, 12(7), e5756.
doi.org/10.1097/GOX.0000000000005756Cleary, M., Edwards, C., Mitchell-Watson, J., Yang, J., & Reddan, T. (2024). Benchmarking non-attendance patterns in pediatric medical imaging: A retrospective cohort study spotlighting First Nations children. Radiography, 30(2), 492-499.
doi.org/10.1016/j.radi.2024.01.002Dantas, L. F., Fleck, J. L., Oliveira, F. L. C., & Hamacher, S. (2018). No-shows in appointment scheduling: A systematic literature review. Health Policy, 122(4), 412-421.
doi.org/10.1016/j.healthpol.2018.02.002Franke, T. M., Ho, T., & Christie, C. A. (2012). The chi-square test: Often used and more often misinterpreted. American Journal of Evaluation, 33(3), 448-458.
doi.org/10.1177/1098214011426594Greenland, S., Senn, S. J., Rothman, K. J., Carlin, J. B., Poole, C., Goodman, S. N., & Altman, D. G. (2016). Statistical tests, P values, confidence intervals, and power: A guide to misinterpretations. European Journal of Epidemiology, 31(4), 337-350.
doi.org/10.1007/s10654-016-0149-3Hess, A. S., & Hess, J. R. (2017). Understanding tests of the association of categorical variables: The Pearson chi-square test and Fisher’s exact test. Transfusion, 57(4), 877-879.
doi.org/10.1111/trf.14057Kim, H. Y. (2017). Statistical notes for clinical researchers: Chi-squared test and Fisher’s exact test. Restorative Dentistry & Endodontics, 42(2), 152-155.
doi.org/10.5395/rde.2017.42.2.152Kovach, J. V., & Borikar, S. (2018). Enhancing financial performance: An application of Lean Six Sigma to reduce insurance claim denials. Quality Management in Health Care, 27(3), 165-171.
doi.org/10.1097/QMH.0000000000000175Lang, T. A., & Altman, D. G. (2015). Basic statistical reporting for articles published in biomedical journals: The SAMPL guidelines. International Journal of Nursing Studies, 52(1), 5-9.
doi.org/10.1016/j.ijnurstu.2014.09.006Lydersen, S., Fagerland, M. W., & Laake, P. (2009). Recommended tests for association in 2 x 2 tables. Statistics in Medicine, 28(7), 1159-1175.
doi.org/10.1002/sim.3531Mander, G. T. W., Reynolds, L., Cook, A., & Kwan, M. M. (2018). Factors associated with appointment non-attendance at a medical imaging department in regional Australia: A retrospective cohort analysis. Journal of Medical Radiation Sciences, 65(3), 192-199.
doi.org/10.1002/jmrs.284McHugh, M. L. (2013). The chi-square test of independence. Biochemia Medica, 23(2), 143-149.
doi.org/10.11613/BM.2013.018Mohammadi, I., Wu, H., Turkcan, A., Toscos, T., & Doebbeling, B. N. (2018). Data analytics and modeling for appointment no-show in community health centers. Journal of Primary Care & Community Health, 9, 1-11.
doi.org/10.1177/2150132718811692Norbash, A., Yucel, K., Yuh, W., Doros, G., Ajam, A., Lang, E. V., Pauker, S., & Mayr, N. (2016). Effect of team training on improving MRI study completion rates and no-show rates. Journal of Magnetic Resonance Imaging, 44(4), 1040-1047.
doi.org/10.1002/jmri.25219Pearson, K. (1900). On the criterion that a given system of deviations from the probable in the case of a correlated system of variables is such that it can be reasonably supposed to have arisen from random sampling. The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science, 50(302), 157-175.
doi.org/10.1080/14786440009463897Rahman, H. A., Noraidi, A. A., Khalid, A. N. H., Mohamad-Adam, A. Z., Zahari, N. H., & Tuming, N. E. (2025). Practical guide to calculate sample size for chi-square test in biomedical research. BMC Medical Research Methodology, 25, 144.
doi.org/10.1186/s12874-025-02584-4Sharpe, D. (2015). Your chi-square test is statistically significant: Now what? Practical Assessment, Research & Evaluation, 20(8), 1-10.
doi.org/10.7275/tbfa-x148Sullivan, G. M., & Feinn, R. (2012). Using effect size, or why the P value is not enough. Journal of Graduate Medical Education, 4(3), 279-282.
doi.org/10.4300/JGME-D-12-00156.1von Elm, E., Altman, D. G., Egger, M., Pocock, S. J., Gøtzsche, P. C., & Vandenbroucke, J. P. (2007). The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement. PLoS Medicine, 4(10), e296.
doi.org/10.1371/journal.pmed.0040296Wasserstein, R. L., & Lazar, N. A. (2016). The ASA’s statement on p-values: Context, process, and purpose. The American Statistician, 70(2), 129-133.
doi.org/10.1080/00031305.2016.1154108CrashCourse. (2018, August 29). Chi-square tests: Crash Course Statistics #29 [Video]. YouTube.
youtube.com/watch?v=7_cs1YlZougPrepared as an executive research and education resource by Kelly Emrick, DHSc, PhD, MBA, BSRT(ARRT)R. The statistical examples are synthetic and should not be used for patient-care, staffing, or financial decisions without validation against local data, governance review, and appropriate statistical consultation.