Abstract
Multiple Imputation (MI) is a popular method for dealing with missing data in applications. A critical step in MI involves using Rubin’s rules to combine statistics from the analysis of imputed data. However, Rubin’s rules require a sufficient condition that the estimator for the parameter of interest θ ∈ ℝk must be (approximately) multivariate normal. This condition is problematic for non-normal statistics. We propose new rules for combining Chi-square statistics (obtained from analysis of each imputed data) by first transforming them to normal, and then combining them using Rubin’s univariate combining rules. Simulation results demonstrate that our proposed methodology controls Type I error rates and provides superior statistical power in small-to-moderate sample sizes (n > k, but not n ≫ k). The proposed method reduces the computational burden from O(k3) to O(1) for inference about θ, and eliminates reliance on the assumption of proportional and equal loss of information, which is assumed in Rubin’s traditional multivariate combining rules to obtain a stable, non-singular variance-covariance matrix. These advantages make the proposed method particularly effective for missing data applications with small-to-moderate sample sizes.
Key words and phrases: Chi-square statistics, Combining rules, Missing at random, Missing data, Multiple imputation, Normal transformation, Proportional and equal loss of information
Information
| Preprint No. | SS-2026-0062 |
|---|---|
| Manuscript ID | SS-2026-0062 |
| Complete Authors | Ashok K Chaurasia, Mohsen Ebadi |
| Corresponding Authors | Ashok K Chaurasia |
| Emails | achaurasia.uconn@gmail.com |
References
- Barnard, J. and D. B. Rubin (1999). Small-sample degrees of freedom with multiple imputation. Biometrika 86, 948–955.
- Bishop, Y. M. (1969). Full contingency tables, logits, and split contingency tables. Biometrics, 383–399.
- Canal, L. (2005). A normal approximation for the chi-square distribution. Computational statistics & data analysis 48(4), 803–808.
- Chan, K. W. (2022). General and feasible tests with multiply-imputed datasets. The Annals of Statistics 50(2), 930–948.
- Chan, K. W. and X.-L. Meng (2022). Multiple improvements of multiple imputation likelihood ratio tests. Statistica Sinica 32(3), 1489–1514.
- Chaurasia, A. (2023). Combining rules for F- and Beta-statistics from multiply-imputed data. Econometrics and Statistics 25, 51–65.
- Goria, M. N. (1992). On the fourth root transformation of chi-square. Australian Journal of Statistics 34(1), 55–64.
- Grund, S., O. L¨udtke, and A. Robitzsch (2016). Pooling anova results from multiply imputed datasets. Methodology.
- Li, K. H., T. E. Raghunathan, and D. B. Rubin (1991). Large sample significance levels from multiply imputed data using moment-based statistics and an f reference distribution. Journal of the American Statistical Association 86, 1065–1073.
- Little, R. and D. Rubin (2002). Statistical analysis with missing data (2 ed.). New York: John Wiley & Sons.
- Meng, X.-L. and D. B. Rubin (1992). Performing likelihood ratio tests with multiply-imputed data sets. Biometrika 79(1), 103–111.
- Peizer, D. B. and J. W. Pratt (1968). A normal approximation for binomial, F, Beta, and other common, related tail probabilities, I. Journal of the American Statistical Association 63(324), 1416–1456.
- Raghunathan, T. and Q. Dong (2011). Analysis of Variance from Multiply Imputed Data Sets. Ann Arbor: University of Michigan.
- Reiter, J. P. (2007). Small-sample degrees of freedom for multi-component significance tests with multiple imputation for missing data. Biometrika 94(2), 502–508.
- Rubin, D. (1987). Multiple Imputation for Nonresponse in Surveys. New York: John Wiley & Sons.
- Schafer, J. (1997). Analysis of Incomplete Multivariate Data. London: Chapman and Hall.
- Van Buuren, S. (2012). Flexible imputation of missing data, Volume 10. CRC press Boca Raton, FL.
- Van Buuren, S. and K. Groothuis-Oudshoorn (2011). mice: Multivariate imputation by chained equations in R. Journal of statistical software 45, 1–67.
- Van-Buuren, S. and C. Oudshoorn (1999). Flexible multivariate imputation by mice. Leiden: TNO Preventie en
- Gezondheid, TNO/VGZ/PG 99.054.
- van Ginkel, J. R. (2019). Significance tests and estimates for R2 for multiple regression in multiply imputed datasets: A cautionary note on earlier findings, and alternative solutions. Multivariate Behavioral Research 54(4), 514– 529.
- Wilson, E. B. and M. M. Hilferty (1931). The distribution of chi-square. Proceedings of the National Academy of Sciences 17(12), 684–688. Ashok Chaurasia
Acknowledgments
The authors thank the Co-Editor, the Associate Editor, and the two anonymous referees for seeing the value of our work in the initial submission and for their insightful comments that helped improve the final manuscript. This research was supported by Discovery Grant from the Natural Science and Engineering Research Council of Canada to Ashok Chaurasia [RGPIN-2018-05479]. The funding agency had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The opinions expressed in this manuscript are those of the authors and do not necessarily reflect the views of their respective institutions.