Introduction to Human Nutrition

Nutrition Research Methodology 309

distribution, can be used. Nonparametric tests are
also useful for data collected as ordinal variables
because they are based on ranking the values. Relative
to their parametric counterparts, nonparametric tests
have the advantage of ease, but the disadvantage of
less statistical power if a normal distribution should
be assumed. Another additional disadvantage is that
they do not permit confi dence intervals for the differ-
ence between means to be estimated.

A common problem in nutrition literature is mul-

tiple signifi cance testing. Some methods to consider

in these instances are analysis of variance together

with multiple-comparison methods specially designed

to make several pairwise comparisons, such as the

least signifi cant difference method, the Bonferroni

and Scheffé procedures, and the Duncan test. Analysis

of variance can also be used for replicate measure-

ments of a continuous variable.

Table 13.3 Common statistical methods used in nutritional epidemiology

Dependent variable (“outcome”) Univariate description Bivariate comparisons Multivariable analysis (Katz, 2006)

Quantitative (normal)

Mean, standard deviation t-Tests (two groups)

Analysis of variance (more

than two groups)

Regression and correlation

(two quantitative

variables)

Multiple regression

Qualitative (dichotomous)

Proportion, odds Chi-squared

McNemar paired test

Fisher’s exact test

Cross-tables

Odds ratio

Relative risk

Multiple logistic regression

Conditional logistic regression

(matched data)

Survival

0 18016014012010080604020

100

90

80

70

60

50

40

30

20

Length of follow-up

Cumulative survival (%)

Kaplan–Meier (product-

limit) estimates and

plots

Log-rank test

(Mantel–Haenszel)

Proportional hazards model (Cox

regression)

Table 13.4 Common statistical methods for comparison of means

Two samples More than two samples

Parametric Nonparametric Parametric Nonparametric

Independent
samples Paired
or related
samples

Student’s t-test

Welch test (unequal variances)

Satterthwaite test (unequal

variances)

Paired t-test

Mann–Whitney U-test

Wilcoxon test

Analysis of variance

Bonferroni, Scheffé, Tamhane, Dunnet,

Sidak, or Tukey post-hoc tests

Analysis of variance for repeated

measurements

General linear models

ANCOVA (analysis of covariance)

Kruskal–Wallis test

Friedman’s test