7 Robustness of ANOVA and MANOVA test procedures

doi:10.1016/S0169-7161(80)01009-7

Handbook of Statistics

Volume 1, 1980, Pages 199-236

https://doi.org/10.1016/S0169-7161(80)01009-7 Get rights and content

Publisher Summary

This chapter discusses the robustness of univariate analysis of variance (ANOVA) and the multivariate analysis of variance (MANOVA) test procedures. A statistical hypothesis in the (ANOVA) and MANOVA is usually tested on the assumption that the observations are (1) independently and (2) normally distributed (3) with a common variance or variance-covariance (var-covar) matrix. A desirable characteristic of a test is that while it is powerful—that is, sensitive to changes in the specified factors under test—it is also robust—that is, insensitive to changes in extraneous factors not under test. Specifically, a test is called robust when its significance level (type-I error probability) and power (one minus type-II error probability) are insensitive to departures from the assumptions on which it is derived. The magnitudes of the effects on ANOVA and MANOVA test procedures are examined when (1) the observations, while remaining independent of one another, are not normally distributed and (2) the observations, while remaining independent of one another, are normally but heteroscedastically distributed—that is, with different variances or varcovar matrices. In most cases, the violations of the assumptions of normality and of homoscedasticity should be treated one at a time.

References (45)

AndersonT.W.
An Introduction to Multivariate Statistical Analysis
(1958)
BartlettM.S.
The effect of non-normality on the t-distribution
BartlettM.S.
A note on tests of significance in multivariate analysis
BoxG.E.P.
Some theorems on quadratic forms applied in the study of analysis of variance problems. I. Effects of inequality of variance in the one-way classification
Ann. Math. Statist.
(1954)
BoxG.E.P.
Some theorems on quadratic forms applied in the study of analysis of variance problems. II. Effects of inequality of variance and of correlation between errors in the two-way classification
Ann. Math. Statist.
(1954)
BoxG.E.P. et al.
Permutation theory in the derivation of robust criteria and the study of departures from assumption
J. Roy. Statist. Soc. B
(1955)
BoxG.E.P. et al.
Robustness to non-normality of regression tests
Biometrika
(1962)
ChaseG.R. et al.
A Monte Carlo investigation of the robustness of T²
J. Amer. Statist. Assoc.
(1971)
DavidF.N. et al.
The effect of non-normality on the power function of the F-test in the analysis of variance
Biometrika
(1951)
DonaldsonT.S.
Robustness of the F-test to errors of both kinds and the correlation between the numerator and denominator of the F-ratio
J. Amer. Statist. Assoc.
(1968)

GayenA.K.

The distribution of ‘Student’ t in random samples of any size drawn from non-normal universes

Biometrika

(1949)

GayenA.K.

The distribution of the variance ratio in random samples of any size drawn from non-normal universes

Biometrika

(1950)

GearyR.C.

The distribution of ‘Student's’ ratio from non-normal samples

J. Roy. Statist. Soc.

(1936)

GearyR.C.

Testing for normality

Biometrika

(1947)

HollowayL.N. et al.

The robustness of Hotelling's T²

J. Amer. Statist. Assoc.

(1967)

HopkinsJ.W. et al.

Some empirical distributions of bivariate T² and homoscedasticity criterion M under unequal variance and leptokurtosis

J. Amer. Statist. Assoc.

(1963)

HorsnellG.

The effect of unequal group variances on the F-test for the homogeneity of group means

Biometrika

(1953)

HotellingH.

A generalized T-test and measure of multivariate dispersion

ItoK.

On the effect of heteroscedasticity and nonnormality upon some multivariate test procedures

ItoK. et al.

On the robustness of the T²₀ test in multivariate analysis of variance when variance-covariance matrices are not equal

Biometrika

(1964)

JamesG.S.

Tests of linear hypotheses in univariate and multivariate analysis when the ratios of the population variances are unknown

Biometrika

(1954)

KendallM.G. et al.

Cited by (134)

A prospective cohort study of familial versus sporadic essential tremor cases: Do clinical features evolve differently across time?
2023, Journal of the Neurological Sciences
Although essential tremor (ET) is often divided into familial and sporadic cases, few data compare the evolution of clinical features in these groups over time. Leveraging data from a prospective, longitudinal study, we present analyses of the evolution of a broad range of cognitive, motor (i.e., tremor, tandem gait) and other features (e.g., disability) of ET.
Sixty-six familial and 23 sporadic ET cases completed in-home evaluations at baseline and 18, 36, and 54-month follow-ups. Assessments included detailed neuropsychological testing and videotaped neurological examinations. Analyses compared the longitudinal course of 16 clinical features in familial and sporadic cases.
Baseline mean age was 75.2 ± 8.8 years and mean observation period was 4.7 ± 0.3 years. Tremor onset age was lower and childhood onset more common in familial than sporadic cases (p's = 0.02). Longitudinal analyses revealed no significant differences between clinical features displayed by familial and sporadic cases, or differences between the patterns of change in clinical features observed in these groups across time. Sporadic cases' daily activity skills declined significantly, whereas familial cases' did not, p's = 0.04 and 0.34, respectively; however, this finding was non-significant when controlling for false discovery rate. Several additional non-significant trends were noted.
Familial and sporadic ET cases differed in onset age, and in the prevalence of childhood tremor onset. Although a number of interesting trends were observed, no significant differences in the evolution of clinical features over time in patients with and without a family history of ET were revealed.
An evaluation of a mental health literacy program for community sport leaders
2023, Mental Health and Prevention
Community sports coaches and leaders can have an important role to play as informal resources for sporting adolescents and adults at risk of mental ill-health. Supporting coach mental health literacy using programs has potential to enhance opportunities for coach-player support. This study aimed to evaluate a community sport mental health literacy program and its capacity to improve club leader confidence to support and refer (primary outcome), awareness of resources, beliefs about mental health and stigmatizing attitudes (social distance). A cluster-randomized controlled evaluation was conducted with measures taken at baseline (around 2 weeks prior to intervention), immediately post-program and at a nine-month follow up. Twenty-five community Australian Rules Football clubs from metropolitan and regional centres were sampled and matched by size and location. Coaches or sport leaders (n = 265) were randomized by club to either the intervention group or a waitlist control group. A multivariate analysis of variance (pre-test scores as covariates) revealed participants in the ‘Tackle Your Feelings’ (TYF) program reported improved outcomes compared to controls on measures tied to leader confidence to support and refer, knowledge of resources to support mental health, general help-seeking and stigmatizing attitudes tied to social distance. A fifth measure of beliefs about mental illness was not impacted by the program. A nine-month comparison group follow-up indicated some decay in scores over time. The results suggest that TYF programs can be effective at improving coach mental health literacy in community sporting club contexts.
Meat-eating justification when gender identity is threatened – The association between meat and male masculinity
2023, Food Quality and Preference
Identifying psychological barriers that prevent individuals from adopting a meat-free diet is a key to success for environmentalists. The current study experimentally investigates the relation between sex, sex role identification, and meat-eating justification by exposing participants to a gender identity threat or affirmation. In the threat condition, participants received the information that they had below-average levels of masculinity/femininity, whereas in the affirmation group they were told to be above average with regard to their masculinity/femininity. Two-hundred female and 132 male participants were randomly assigned to these conditions. The findings suggest that men and women rationalize their meat consumption differently: While men directly and unapologetically defend their behavior, women tend to avoid recognizing their own responsibility. Moreover, masculinity was associated with direct meat-eating justifications among male, but not among female participants. The experimental manipulation did not affect direct or indirect meat-eating justification. However, further exploratory analyses suggest that how threatening the manipulated feedback was perceived was not primarily dependent on the participant’s sex, but rather the individual identification with sex roles. Therefore, simply basing gender identity threats on participant’s sex might be outdated. The results highlight the association between meat and male masculinity and encourage future studies to take a more nuanced approach to gender identity threats.
An algorithm for simulation in mixed models with crossed factors considering the sample sizes as random
2022, Journal of Computational and Applied Mathematics
Applying analysis of variance (ANOVA) when the sample dimensions are not known in advance is not such a rare situation. These situations frequently occur when a fixed time period is established for collecting the observations. In consequence, the number of observations that are obtained for each treatment is not fixed, depending on the arrival rate. So, since we cannot anticipate the exact sample dimensions, we should consider them as realizations of independent random variables. Therefore, the aim of this research is to extend the theory of the orthogonal mixed models to those situations, assuming the sample sizes as Poisson distributed. The model formulation is done considering stable statistics, which means that the test statistics, under the null hypothesis, have the same distribution whether referring to the fixed or to the random effects part of the model. The approach is extended to a general model with crossing of several factors. The applicability of the proposed approach is illustrated through an application based on real data, considering the incidence of unemployed persons in the European Union. The obtained results suggest that false rejections may be avoided when applying our approach, which is confirmed by carrying out two simulation studies.
Collective behaviour of the European minnow (Phoxinus phoxinus) is influenced by signals of differing acoustic complexity
2021, Behavioural Processes
Collective behaviour, such as shoaling in fish, benefits individuals through a variety of activities such as social information exchange and anti-predator defence. Human driven disturbance (e.g. anthropogenic noise) is known to affect the behaviour and physiology of individual animals, but the disruption of social aggregations of fish remains poorly understood. Anthropogenic noise originates from a variety of activities and differs in acoustic structure, dominant frequencies, and spectral complexity. The response of groups of fish may differ greatly, depending on the type of noise, and how it is perceived (e.g. threatening or attractive). In a controlled laboratory study, high resolution video tracking in combination with fine scale acoustic mapping was used to investigate the response of groups of European minnows (Phoxinus phoxinus) to signals of differing acoustic complexity (sinewave tones vs octave band noise) under low (150 Hz) and high (2200 Hz) frequencies. Fish startled and decreased their mean group swimming speed under all four treatments, with low frequency sinewave tones having the greatest influence on group behaviour. The shoals exhibited spatial avoidance during both low frequency treatments, with more time spent in areas of lower acoustic intensity than expected. This study illustrates how noise can influence the spatial distribution and social dynamics within groups of fish, and owing to the high potential for freshwater aquatic environments to be influenced by anthropogenic activity, wider consequences for populations should be further investigated.
Exploring the possibility of using CO<inf>2</inf> as a proxy for exhaled particles to predict the risk of indoor exposure to pathogens
2023, Indoor and Built Environment

View all citing articles on Scopus

View full text

7 Robustness of ANOVA and MANOVA test procedures

Publisher Summary

An Introduction to Multivariate Statistical Analysis

The effect of non-normality on the t-distribution

A note on tests of significance in multivariate analysis

Some theorems on quadratic forms applied in the study of analysis of variance problems. I. Effects of inequality of variance in the one-way classification

Ann. Math. Statist.

Some theorems on quadratic forms applied in the study of analysis of variance problems. II. Effects of inequality of variance and of correlation between errors in the two-way classification

Ann. Math. Statist.

Permutation theory in the derivation of robust criteria and the study of departures from assumption

J. Roy. Statist. Soc. B

Robustness to non-normality of regression tests

Biometrika

A Monte Carlo investigation of the robustness of T2

J. Amer. Statist. Assoc.

The effect of non-normality on the power function of the F-test in the analysis of variance

Biometrika

Robustness of the F-test to errors of both kinds and the correlation between the numerator and denominator of the F-ratio

J. Amer. Statist. Assoc.

The distribution of ‘Student’ t in random samples of any size drawn from non-normal universes

Biometrika

The distribution of the variance ratio in random samples of any size drawn from non-normal universes

Biometrika

The distribution of ‘Student's’ ratio from non-normal samples

J. Roy. Statist. Soc.

Testing for normality

Biometrika

The robustness of Hotelling's T2

J. Amer. Statist. Assoc.

Some empirical distributions of bivariate T2 and homoscedasticity criterion M under unequal variance and leptokurtosis

J. Amer. Statist. Assoc.

The effect of unequal group variances on the F-test for the homogeneity of group means

Biometrika

A generalized T-test and measure of multivariate dispersion

On the effect of heteroscedasticity and nonnormality upon some multivariate test procedures

On the robustness of the T20 test in multivariate analysis of variance when variance-covariance matrices are not equal

Biometrika

Tests of linear hypotheses in univariate and multivariate analysis when the ratios of the population variances are unknown

Biometrika

A Monte Carlo investigation of the robustness of T²

The robustness of Hotelling's T²

Some empirical distributions of bivariate T² and homoscedasticity criterion M under unequal variance and leptokurtosis

On the robustness of the T²₀ test in multivariate analysis of variance when variance-covariance matrices are not equal