Center for Global Health

1 University of New Mexico
Albuquerque, NM 87131

Phone: (505) 272-8207
Fax: (505) 272-8441

Laboratories of Tropical Diseases Biostatistics

Dr. Jim Hittner

Dr. Jim Hittner
Director of
Statistical Methodology

The site of research and training in the USA is located in the University of New Mexico, Department of Internal Medicine, Division of Infectious Diseases. Research and training facilities for the Kenyan activities are located at the University of New Mexico Laboratories of Parasitic and Viral Diseases in Kisumu and Siaya Districts.

To understand the complex interplay between immunological parameters, malaria-related morbidity, and associated co-infections (e.g., bacteremia, HIV), we utilize a spectrum of inferential statistical techniques. Many of the techniques that we use are classical general linear modeling strategies such as multiple linear regression analysis, logistic regression analysis, analysis of covariance, simple correlation (Pearson, partial, semipartial), and discriminant function analysis.

For those situations in which parametric distributional assumptions are not met, we conduct nonparametric rank-based analyses such as Kruskal-Wallis analyses of variance and Mann-Whitney U-tests. Although rank-based approaches typically are appropriate for small sample size analyses, for very small sample sizes (e.g., n ‹ about 10-15), traditional nonparametric techniques often incur low statistical power. For this reason, when we conduct subsample analyses on very small groups of children, we perform randomization tests as a complement to traditional nonparametric procedures.

Randomization tests, also known as permutation tests, are a type of resampling-based analysis. In particular, when conducting a randomization test, the following steps are performed:

  1. a test statistic, such as a t-test, is calculated on the observed data,
  2. random samples (without replacement) for each group are drawn from the observed data, and the test statistic is recalculated on the randomly sampled data,
  3. the process of selecting random samples and recalculating the test statistic is repeated a large number of times (e.g., 10000 times).

The proportion of the recalculated test statistics that are greater than or equal to the statistic obtained on the observed data represents the empirical probability value, or p-value. Simulation studies have shown that randomization tests maintain greater statistical power than do traditional nonparametric techniques with very small sample sizes.

In addition to the challenges associated with analyzing very small subsamples, another important challenge is to develop and test appropriate models for prospective, longitudinal data. One such strategy is to construct multivariable path models depicting the temporal associations among the variables of interest, and to test the statistical fit of such models via multivariable path analysis. The overall fit of such models and the magnitudes of the various direct and indirect effects are instrumental in understanding the nature and patterning of associations over time. Our group has conceptualized several important multivariable path models that will be subjected to testing. In addition to path analysis, we also perform mixed-model repeated-measures analyses and survival analyses to examine change over time and mortality outcomes, respectively.