ArticlePDF Available

Contrast analysis: A tutorial

Authors:
Antal Haans at Eindhoven University of Technology
Antal Haans
Download full-text PDF
Read full-text
Download citation

Abstract

Contrast analysis is a relatively simple but effective statistical method for testing theoretical predictions about differences between group means against the empirical data. Despite its advantages, contrast analysis is hardly used to date, perhaps because it is not implemented in a convenient manner in many statistical software packages. This tutorial demonstrates how to conduct contrast analysis through the specification of the so-called L (the contrast or test matrix) and M matrix (the transformation matrix) as implemented in many statistical software packages, including SPSS and Stata. Through a series of carefully chosen examples, the main principles and applications of contrast analysis are explained for data obtained with between- and within-subject designs, and for designs that involve a combination of between- and within-subject factors (i.e., mixed designs). SPSS and Stata syntaxes as well as simple manual calculations are provided for both significance testing and contrast-relevant effect sizes (e.g., η² alerting). Taken together, the reader is provided with a comprehensive and powerful toolbox to test all kinds of theoretical predictions about cell means against the empirical data.
Content uploaded by Antal Haans
Author content
All content in this area was uploaded by Antal Haans on May 23, 2022
Content may be subject to copyright.
Haans,A.(2018).ContrastAnalysis:ATutorial.PracticalAssessment,Research,andEvaluation,
23,Article9.
Pleasefindthearticleonthewebsiteofthepublisher(openaccess):
https://doi.org/10.7275/7dey‐zd62
... Daran anschließend wurden geplante Interaktionskontraste berechnet, die Auskunft darüber gaben, ob es entsprechend den Unterhypothesen zwischen bestimmten AZ-Formen Unterschiede bezüglich des Ausmaßes an Befriedigung im Vergleich zum Ausmaß an Frustration gab bzw. nicht gab (Haans, 2018;Howells & Lacroix, 2012). ...
... Zur Überprüfung der Hypothesen zu Unterschieden zwischen den AZ-Formen bezüglich der Verwendung der im EVLN-Ansatz unterschiedenen Strategien wurde zunächst zu jeder der fünf Strategien eine Varianzanalyse mit der AZ-Form als Zwischensubjektfaktor durchgeführt. Daran anschließend wurden geplante Kontraste berechnet, die Auskunft darüber gaben, ob es hinsichtlich der einzelnen Strategien den Unterhypothesen entsprechende Unterschiede zwischen bestimmten AZ-Formen gab (Haans, 2018;Howells & Lacroix, 2012). Die Varianzanalysen ergaben, wie erwartet (Hypothesen 4 bis 8), Unterschiede zwischen den fünf AZ-Formen hinsichtlich aller fünf Strategien (s. ...
Formen der Arbeitszufriedenheit sensu Bruggemann: Ein Test der Annahmen des ‚Zürcher Modells‘ zur Rolle des Soll-Ist-Vergleichs, der Regulation des Anspruchsniveaus und den Problemlösungsversuchen (finally)
Article
Full-text available
  • Apr 2024
Zusammenfassung: Im Modell der Arbeitszufriedenheit von Bruggemann (1974 ) werden Bedürfnisse und Erwartungen bezüglich der Arbeit sowie Problemlöseversuche als wesentliche Faktoren für die Entstehung verschiedener Formen der Arbeits‍(un)‌zufriedenheit (fixierte Unzufriedenheit, konstruktive Unzufriedenheit, resignative Zufriedenheit, stabilisierte Zufriedenheit, progressive Zufriedenheit) postuliert. Wir haben die Annahmen des Modells in einer Studie mit N = 611 Teilnehmenden überprüft. Dabei zeigten sich zum einen größtenteils hypothesenkonforme Unterschiede zwischen den via Selbstzuordnungsmethode ermittelten Formen der Arbeitszufriedenheit (AZ-Formen) bezüglich des Ausmaßes an Befriedigung im Vergleich zum Ausmaß an Frustration der in der Selbstbestimmungstheorie ( Ryan & Deci, 2000 ) postulierten Bedürfnisse nach Autonomie, Kompetenz und Eingebundenheit. Zum anderen wurden anhand der EVLN-Typologie ( Hagedoorn, Van Yperen, Van de Vliert & Buunk, 1999 ) modellkonforme Unterschiede zwischen den AZ-Formen im Umgang mit Problemen bei der Arbeit ermittelt.
View
... We evaluated the model-estimated differences in end-exposure protein concentrations between the 22 °C indoor exposure simulation and the other three conditions (26 °C, 31 °C, and 36 °C). We also examined the linear trend between end-exposure protein concentrations and each exposure condition to estimate the slope of the relation between the end-exposure protein concentrations and ambient temperature (Rosner 2015;Haans 2019;Mulder 2022). ...
Brain-derived neurotrophic factor in older adults exposed to simulated indoor overheating
Article
Full-text available
  • Oct 2024
  • EUR J APPL PHYSIOL
Purpose Brain-derived neurotrophic factor (BDNF) is a neuroprotective growth factor that increases in young adults during short, intense bouts of passive heat stress. However, this may not reflect the response in heat-vulnerable populations exposed to air temperatures more consistent with indoor overheating during hot weather and heatwaves, especially as the BDNF response to acute stressors may diminish with increasing age. We therefore evaluated the ambient and body temperature-dependent responses of BDNF in older adults during daylong passive heating. Methods Sixteen older adults (6 females; aged 66–78 years) completed 8-h exposure to four randomized ambient conditions simulating those experienced indoors during hot weather and heatwaves in continental climates: 22 °C (air-conditioning; control), 26 °C (health-agency-recommended indoor temperature limit), 31 °C, and 36 °C (non-airconditioned home); all 45% relative humidity. To further investigate upstream mechanisms of BDNF regulation during thermal strain, we also explored associations between BDNF and circulating heat shock protein 70 (HSP70; taken as an indicator of the heat shock response). Results Circulating BDNF was elevated by ~ 28% (1139 [95%CI: 166, 2112] pg/mL) at end-exposure in the 36 °C compared to the 22 °C control condition (P = 0.026; 26 °C—and 31 °C—22 °C differences: P ≥ 0.090), increasing 90 [22, 158] pg/mL per 1 °C rise in ambient temperature (linear trend: P = 0.011). BDNF was also positively correlated with mean body temperatures (P = 0.013), which increased 0.12 [0.10, 0.13]°C per 1 °C rise in ambient temperature (P < 0.001). By contrast, serum HSP70 did not change across conditions (P ≥ 0.156), nor was it associated with BDNF (P = 0.376). Conclusion Our findings demonstrate a progressive increase in circulating BDNF during indoor overheating in older adults.
View
... With contrast analyses, we directly test the prediction that the effect of wasting food on social status is attenuated if information about considerateness exists (Haans, 2018). As predicted, social status perceptions decreased significantly more when Jack wasted the spaghetti (cf. ...
View
... For the varying conditions, i.e., RW0%, RW5%, RW10% and CONT, contrast analysis (Haans, 2008;Hervé and Williams, 2010) were carried out to specifically test the following hypotheses; H1) RW10% would lead to greater changes in the RPE scores and 20-m sprint performance compared with all other conditions (RW0%, RW5%, CONT), H2). RW10% would yield greater improvements in the RPE scores and 20-m sprint performance compared to RW0% and RW5%, H3) RW10% would yield greater improvements in the RPE scores and 20-m sprint performance when compared with CONT. ...
A 3-min weighted vests re-warmups induce sprint performance enhancements at the start of the second half of a soccer match-play
Article
Full-text available
  • Jul 2023
Purpose: This study aimed to investigate the effect of re-warm-up (RWUP) activities consisting of three sets of 15-m sprints with weighted vests on 20-m sprint performance after the break. Methods: Using a randomized, and counterbalanced cross-over design, twenty U15 elite soccer players performed four RWUP trials which consisted of 15-min of passive rest (control: CONT), 3 × 15-m sprint (RW0%), sprinting with a vest-loaded at 5% of body-mass (BM) (RW5%), sprinting with a 10% BM vest (RW10%). The 3-min RWUP protocols started 10-min after the commencement of the 15-min between-halves break and concluded 2-min before its end. During each testing session, the participant’s RPE scores, and 20-m sprint performance were registered before the first half, and at the end of the break of the simulated match-play. Results: Significant time effects [F (1.72) = 54.37, d = 1.88, p < 0.001; F (1.72) = 9.07, d = 0.77, p = 0.003], and condition effects [F (3.72) = 11.81, d = 1.53, p < 0.001; F (3.72) = 7.36, d = 1.21, p = 0.003] were observed for RPE scores and 20-m sprint performance, respectively. Significant condition-by-time interactions were found for RPE scores (p < 0.003, d = 0.54), and 20-m sprint performance (p < 0.002, d = 0.70). Contrast analysis showed significantly higher RPE scores (p < 0.001, d = 1.55), and improved 20-m sprint performance (p = 0.0004, d = 1.19) in the RW10% condition compared to all other conditions. Conclusion: Sprinting for 3-min with a 10% body-mass vest resulted in the highest RPE scores and the most substantial enhancements in 20-m sprint performance. Young elite soccer players should incorporate 10% body-mass weighted vests in their re-warm-ups to boost post-break sprint performance.
View
... Therefore, dummy plots were included in the statistical analysis. As a result, to examine the effect of plant roots presence on soil physical attributes, total data were classified as two groups (i.e., unplanted vs. planted) and statistical comparisons between unplanted vs. planted treatments were assessed by orthogonal contrast procedure (Haans, 2018). The effects of S. indica colonization and wheat cultivar on shoot dry weight, SOC, HWSC, fitting parameters of SWRC, soil water retention indices (measured θ s , θ FC and θ PWP ), PAW, E I(PAW) , W I(PAW) , and Dexter's physical quality indices (S 1 and S 2 ) were examined by a factorial arrangement. ...
View
... suggesting the absence of any other systematic effects influencing the four observed WTPs. The combination of a significant and negative contrast statistic and a non-significant residual result, that is, the canonical outcome (Abelson & Prentice, 1997), is strong evidence against the theoretical contrast of interest (Haans, 2018). It appears that choosers' selfenhancement bias trumped their rational decision-making. ...
Strategic thinking in the shadow of self-enhancement: Benefits and costs
Preprint
  • Dec 2022
Using a variant of the hide-and-seek game, we show in two studies that self-enhancement can help or hinder strategic thinking. In the game, players either choose a number or attempt to guess that number. They can do so by randomizing or by active thinking. Guessers profit from thinking of a number, whereas choosers suffer from it. Yet, irrespective of their role, respondents prefer to actively think of a number when the opponent thinks. Participants believe they can outthink an opponent and most even prefer to play the game in the chooser’s role, thus neglecting the guesser’s statistical advantage. These findings suggest that the impacts of self-enhancement on strategic thinking are robust, presenting both risks (for the chooser) and opportunities (for the guesser). We discuss the implications of these findings for theories of strategic cognition and applications to real-life contexts.
View
... Hence, instead of using a contrast analysis, context of participants was tested as such by comparing separate GAM models. Furthermore, the contrast analysis method described by Haans (2018) in Stage 1 turned out to be incompatible with GAM. Therefore, the alternative procedure as described in the statistical analyses was used. ...
Adolescence as a peak period of borderline personality features? A meta-analytic approach
Article
Full-text available
  • Oct 2022
This meta-analysis of cross-sectional data aimed to shed light on the often assumed peak in mean-level of borderline personality features during middle to late adolescence (i.e. age 17–22). Borderline personality features were operationalized through the Structured Clinical Interview for DSM-IV Axis II Personality Disorders (SCID-II). Search terms were entered into PsycINFO and Scopus. A total of 168 samples were included in the analyses, comprising 25,053 participants. Mean age ranged from 14.35 to 51.47 years ( M = 29.01, SD = 8.52) and mean number of borderline personality features from 0 to 8.10 ( M = 4.59, SD = 2.34). The hypothesized peak between age 17 and 22 was not substantiated by the confirmatory ANOVA analysis. However, subsequent exploratory GAM analysis provided evidence for a peak at 29.4 years. Caution is needed in interpreting these findings given that different trends appeared when GAM models were constructed separately for community, patient and borderline personality disorder (BPD) samples. Age differences in community samples indicated a significant linear decline in mean-level of borderline personality features over time. A linear rising trend was found in BPD samples. As a between-person mean-level approach was used in the current study, future longitudinal studies are needed to substantiate if between-person age difference generalize to within-person changes.
View
Effects of Daylong Exposure to Indoor Overheating on Thermal and Cardiovascular Strain in Older Adults: A Randomized Crossover Trial
Article
  • Feb 2024
  • ENVIRON HEALTH PERSP
Background: Health agencies recommend that homes of heat-vulnerable occupants (e.g., older adults) be maintained below 24-28°C to prevent heat-related mortality and morbidity. However, there is limited experimental evidence to support these recommendations. Objective: To aid in the development of evidence-based guidance on safe indoor temperatures for temperate continental climates, we evaluated surrogate physiological outcomes linked with heat-related mortality and morbidity in older adults during simulated indoor overheating. Methods: Sixteen older adults [six women; median age: 72 y, interquartile range (IQR): 70-73 y; body mass index: 24.6 (IQR: 22.1-27.0) kg/m2] from the Ottawa, Ontario, Canada, region (warm summer continental climate) completed four randomized, 8-h exposures to conditions experienced indoors during hot weather in continental climates (e.g., Ontario, Canada; 64 participant exposures). Ambient conditions simulated an air-conditioned environment (22°C; control), proposed indoor temperature upper limits (26°C), and temperatures experienced in homes without air-conditioning (31°C and 36°C). Core temperature (rectal) was monitored as the primary outcome; based on previous recommendations, between-condition differences >0.3°C were considered clinically meaningful. Results: Compared with 22°C, core temperature was elevated to a meaningful extent in 31°C [+0.7°C; 95% confidence interval (CI): 0.5, 0.8] and 36°C (+0.9°C; 95% CI: 0.8, 1.1), but not 26°C (+0.2°C, 95% CI: 0.0, 0.3). Increasing ambient temperatures were also associated with elevated heart rate and reduced arterial blood pressure and heart rate variability at rest, as well as progressive impairments in cardiac and blood pressure responses to standing from supine. Discussion: Core temperature and cardiovascular strain were not appreciably altered following 8-h exposure to 26°C but increased progressively in conditions above this threshold. These data support proposals for the establishment of a 26°C indoor temperature upper limit for protecting vulnerable occupants residing in temperate continental climates from indoor overheating. https://doi.org/10.1289/EHP13159.
View
Information, Assessment, or Decision: A Driving Simulator Study on the Effect of Real-Time Feedback Based on Information-Processing Stages
Preprint
Full-text available
  • Jul 2023
This study, focused on supporting drivers through feedback rather than automating the driving task, examined the effect of real-time feedback based on different stages of information processing on driving behaviour. Using a driving simulator with 29 participants, the stages investigated included information alone, assessment of that information, and a decision based on that assessment, following Parasuraman et al.’s (2000) model of information processing. The acceptability and effectiveness of different stages of feedback on two key driving behaviours: speed and distance from the vehicle ahead, were assessed. The results indicated that feedback alone had a limited effect on driving behaviour. However, the stage of information processing in the feedback did have an effect on outcomes, with decision-oriented feedback leading to improved behavioural outcomes but less favourable attitudinal results. Future safety interventions should consider altering risk perception and beliefs or providing external motivation for behavioural change.
View
Open multi-session and multi-task EEG cognitive Dataset for passive brain-computer Interface Applications
Article
Full-text available
  • Feb 2023
Brain-Computer Interfaces and especially passive Brain-Computer interfaces (pBCI), with their ability to estimate and monitor user mental states, are receiving increasing attention from both the fundamental research and the applied research and development communities. Testing new pipelines and benchmarking classifiers and feature extraction algorithms is central to further research within this domain. Unfortunately, data sharing in pBCI research is still scarce. The COG-BCI database encompasses the recordings of 29 participants over 3 separate sessions with 4 different tasks (MATB, N-Back, PVT, Flanker) designed to elicit different mental states, for a total of over 100 hours of open EEG data. This dataset was validated on a subjective, behavioral and physiological level, to ensure its usefulness to the pBCI community. Furthermore, a proof of concept is given with an example of mental workload estimation pipeline and results, to ensure that the data can be used for the design and evaluation of pBCI pipelines. This body of work presents a large effort to promote the use of pBCIs in an open science framework.
View
Performing Contrast Analysis in Factorial Designs
Article
Full-text available
  • Oct 2016
  • EDUC PSYCHOL MEAS
Because of the continuing debates about statistics, many researchers may feel confused about how to analyze and interpret data. Current guidelines in psychology advocate the use of effect sizes and confidence intervals (CIs). However, researchers may be unsure about how to extract effect sizes from factorial designs. Contrast analysis is helpful because it can be used to test specific questions of central interest in studies with factorial designs. It weighs several means and combines them into one or two sets that can be tested with t tests. The effect size produced by a contrast analysis is simply the difference between means. The CI of the effect size informs directly about direction, hypothesis exclusion, and the relevance of the effects of interest. However, any interpretation in terms of precision or likelihood requires the use of likelihood intervals or credible intervals (Bayesian). These various intervals and even a Bayesian t test can be obtained easily with free software. This tutorial reviews these methods to guide researchers in answering the following questions: When I analyze mean differences in factorial designs, where can I find the effects of central interest, and what can I learn about their effect sizes?
View
Decomposing interactions using GLM in combination with the COMPARE, LMATRIX and MMATRIX subcommands in SPSS
Article
Full-text available
  • Feb 2012
In this tutorial, we provide researchers who use SPSS step-by-step instructions for decomposing interactions when a three-way ANOVA is conducted using the GLM procedure. We start with a demonstration of how a two-way interaction can be decomposed using the COMPARE subcommand in combination with syntax. Then, we provide instructions with examples for conducting simple interaction and second-order simple effects analyses for three-way ANOVAs with between-subjects, within-subjects, and mixed between- and within-subjects variables using the LMATRIX or MMATRIX subcommands. Provided in Appendices are general rules that can be used to derive design-specific LMATRIX and MMATRIX subcommands.
View
What does it mean to be average? The miles per gallon versus gallons per mile paradox revisited
Article
Full-text available
  • Jan 2008
In the efficiency paradox, which was introduced by Hand (1994; J R Stat Soc A, 157, 317-356), two groups of engineers are in disagreement about the average fuel efficiency of a set of cars. One group measured efficiency on a miles per gallon scale, the other on a gallons per mile scale. In the present paper, I argue against an operationalistic explanation of the efficiency paradox, by showing that the paradox is neither the result of an ambiguously defined efficiency concept, nor the result of how fuel efficiency is measured (i.e., whether a miles per gallon, or gallons per mile scale is used). The actual paradox is that the two groups of engineers have asked different statistical questions, while using the same mathematical operation. The paradox results from the fact that fuel efficiency is a derived measure, like density and speed, for which end-to-end concatenation (i.e., addition) is not straightforward.
View
Contrast Tests of Interaction Hypotheses
Article
Full-text available
  • Dec 1997
This article argues for the use of contrasts to test a priori interaction hypotheses in 2-way analysis of variance designs. It focuses on 3 underused types of interaction contrast tests: a "matching" pattern for cognate levels of row and column factors; the "qualitative quadratic" for monotonic profiles of means in the same direction but with opposed concavities; and a "differential curvilinearity" test for differences in the curvature of two profiles with the same direction of concavity. The circumstances that best capitalize on the potential advantages of a priori contrast tests are indicated and an effect size measure for contrasts is presented. Investigators are urged to examine residuals after accounting for the variation attributable to the chosen contrast for patterns that may provide hints for more textured hypotheses in further research. If a posteriori contrasts are used, their effect sizes should be noted. (PsycINFO Database Record (c) 2012 APA, all rights reserved)
View
Things I learned (so far)
Article
Full-text available
  • Dec 1990
This is an account of what I have learned (so far) about the application of statistics to psychology and the other sociobiomedical sciences. It includes the principles "less is more" (fewer variables, more highly targeted issues, sharp rounding off), "simple is better" (graphic representation, unit weighting for linear com- posites), and "some things you learn aren't so." I have learned to avoid the many misconceptions that surround Fisherian null hypothesis testing. I have also learned the importance of power analysis and the determination of just how big (rather than how statistically significant) are the effects that we study. Finally, I have learned that there is no royal road to statistical induction, that the informed judgment of the investigator is the crucial element in the interpretation of data, and that things take time.
View
View
Deconstructing Statistical Questions
Article
  • Jan 1994
Too much current statistical work takes a superficial view of the client's research question, adopting techniques which have a solid history, a sound mathematical basis or readily available software, but without considering in depth whether the questions being answered are in fact those which should be asked. Examples, some familiar and others less so, are given to illustrate this assertion. It is clear that establishing the mapping from the client's domain to a statistical question is one of the most difficult parts of a statistical analysis. It is a part in which the responsibility is shared by both client and statistician. A plea is made for more research effort to go in this direction and some suggestions are made for ways to tackle the problem.
View
Contrasts and Interactions Redux: Five Easy Pieces
Article
  • Jul 1996
This reply to Abelson (this issue) and Petty, Fabrigar, Wegener, and Priester (this issue) is couched within the framework of five basic principles advising that we (1) hang on to what we predicted long enough to test it (Tarzan's leap), (2) be wary of beguiling statistical designs that may not address the question of interest (the Sirens' song), (3) not allow well-worn habits of thinking to ensnare our perceptions (Lavoisier's crease), (4) weigh the possibility of more than one correct hypothesis (the dayyan's decree), and (5) not confuse unplanned with planned contrasts (the archer's aim)
View
“Some Things You Learn Aren't So”: Cohen's Paradox, Asch's Paradigm, and the Interpretation of Interaction
Article
  • Jan 1995
When interpreting an interaction in the analysis of variance (ANOVA), many active researchers (and, in turn, students) often ignore the residuals defining the interaction Although this problem has been noted previously, it appears that many users of ANOVA remain uncertain about the proper understanding of interaction effects To clear up this problem, we review the way in which the ANOVA model enables us to take apart a table of group means or the individual measurements contributing to the means to reveal the underlying components We also show how (using only published data) to compute a contrast on the question that may be of primary interest and illustrate strategies for interpreting tables of residuals We conclude with an exercise to check on students' understanding of ANOVA and to encourage increased precision in the specification of research results
View
Analysis of Variance Frameworks in Clinical Child and Adolescent Psychology: Issues and Recommendations
Article
  • Apr 2002
Reviewed existing practices of factorial analysis of variance (ANOVA), a major analytic tool used in clinical child and adolescent psychology, in the Journal of Clinical Child Psychology (JCCP) and noted several suboptimal strategies. Issues surrounding the analysis of multiple outcome variables, omnibus F tests, and single degree of freedom contrasts, simple main effects analysis, and single degree of freedom interaction contrasts were considered and recommendations were made about analytic strategies. Among the practices questioned were the use of multivariate analyses of variance (MANOVAs) as a means of controlling Type I errors across multiple outcome variables and the use of simple main effects analysis to elucidate the nature of interaction effects.
View