Summary and study notes

Welke onderwerpen worden behandeld in het hoorcollege?

Repeated measures analysis. The same outcome variable is measured multiple times in the same people. 

Mixed design. This design has two factors: within factor and between factor. In mixed designs there can be an interaction effect. 

Assumption of sphericity = = variances of all difference scores are equal. The test for sphericity is called the Mauchly’s test. 

Welke onderwerpen worden besproken die niet worden behandeld in de literatuur?

In dit college worden geen andere onderwerpen besproken dit niet worden behandeld in de literatuur.  

Welke recente ontwikkelingen in het vakgebied worden besproken? 

Er worden geen recente ontwikkelingen besproken. 

Welke opmerkingen worden er tijdens het college gedaan door de docent met betrekking tot het tentamen?

Er worden geen opmerkingen gedaan die betrekking hebben tot het tentamen. 

Welke vragen worden behandeld die gesteld kunnen worden op het tentamen? 

Er worden geen tentamenvragen behandeld. 

Hoorcollege aantekeningen 

When you have an outcome variable which is measured multiple times, then we speak of a repeated measures analysis or a mixed design.

• Within subjects’ factor = the same outcome variable is measures multiple times within a person. 
• Between subjects’ factor = categorical variable which differs per person. 

Designs with one or more within factors are called repeated measures designs.

Designs with at least one between and at least one within factor are formally called mixed designs (but often also repeated measures). Note: all analyses in Firk et al. are done controlling for IQ mother (the covariate): mixed ANCOVA. 

Repeated measures analysis (RMA)

The same outcome variable is measured multiple times in the same people. For example, at different occasions (longitudinal research) or in different conditions (experimental research). It is not always possible to do a within factor analysis because you don’t get each person in each condition. Advantages of within subjects’ design:

• It is more economical (less respondents needed)
• It is more powerful (less noise of unmeasured individual differences)
• Possible to investigate change over time (in longitudinal context)

Dependency of data across conditions requires analysis technique that takes that into account (this is a assumption). Why not a regular ANOVA? Because we know how the data is collected, we know that the means are dependent. So, we know to do a repeated measures analysis. 

Homogeneity of variance (another assumption) is tested with the assumption of sphericity = variances of all difference scores are equal. The test for sphericity is called the Mauchly’s test. Test results are sensitive to sample size, always also inspect descriptive information on size of violation. If you have a really large sample, even the smallest violations become significant. How severe the sphericity is, can you find under Epsilon. The severity is a number between 0 and 1. Hereby, 1 means perfect sphericity. The lower bound tells you the minimum score of Epsilon. 

If sphericity is violated, you need to know when to choose which option:

• Sphericity assumed – no significant Mauchly’s test. 
• Greenhouse-Geisser – really strict, for bigger violations. When epsilon is < .75. 
• Huynh-Feldt – mild, for small violations. When epsilon is > .75.  
• MANOVA – if you don’t trust the sphericity. When epsilon is very low (closer to lower bound than to 1). 

When your result is significant, so there is an effect, you have to investigate what the difference is. This can be done by a plot but can be confusing. Use effect sizes to say how big the effect is. 

When you reject the null hypothesis there is a main effect. After that, you do follow up tests, two options:

1. Post-hoc comparisons between different time points.
2. Planned comparisons: testing pre-specified contrasts (contrast testing). In longitudinal design, often question like “is there a linear increase or decrease over time”, stated differently “is the contrast measuring the linear effect significantly non-zero?”. Or “is there a quadratic pattern over time?”. 

To measure the linear trend in p measurements, we add up the means with (contrast) weights for all means, complying to: the sum of all weights is zero (rule for every contrast) and the distance between the weight is equal (rule for linear contrast). 

Mixed design

This design has two factors: within factor and between factor. In mixed designs there can be an interaction effect. There is also in this design the assumption of sphericity. If you have only two measurements, then you have no assumption of sphericity, so you need minimum three levels. When you know which effects are significant, you know there is a difference (or not). Then you have to look which groups differ from each other. This can by making a plot or by follow up tests: single effect test, post hoc or planned contrast.