5PAHPRM4: Research Methods 4 notes

RM4 LECTURE 1: ANOVA and ANCOVA

1.1: Understand the Types of translational validity:
concepts of reliability and  Face validity= does the measurement technique look like it measures variable it claims to
validity (P + R) measure?
 Construct validity= does the measurement of a variable behave in same way as variable itself?

Types of criterion validity:
 Concurrent validity= are scores obtained from new measurement related to scores obtained
from another, better-established procedure for measuring same variable (i.e. golden standard)?
 Predictive validity= does measurement accurately predict bhvr that, in theory, should relate to
measured variable?
 Convergent validity= is there strong relationship between measurement + other measurements
of same variable?
 Divergent/discriminant validity= does measurement lack a relationship w/constructs it’s not
theoretically related to?
1.4: Understand the use ANOVA= used to evaluate whether two or more groups differ from each other on a single continuous)
(and abuse) of the Analysis ratio or interval) dependent variable
of Covariance (ANCOVA) (P  ANOVA used to test how likely it is these differences occur by chance if we assume groups are
+ L + R) from same underlying pop
- Involves categorical IV – uses between-subjects + involves continuous DV
- One-way ANOVA= only single independent variable, w/several lvls
- Factorial ANOVA= multiple independent variables, each w/several lvls

One-way ANOVA (between) Researchers randomly assign minimal group membership (red vs.
blue vs. yellow), then measure group behaviour.
One-way ANOVA (within) Self-esteem is measured before and after an intervention.
Two-way ANOVA (between) Researchers test if manipulated ostracism (vs. control) increases
anxiety more among adolescents than adults.




EXAMPLE RQ: How does ‘level of processing’ affect learning? 
 50 ppts assigned to either 1 of 5 groups— 
1. Counting group simply counted letters in word simply counted letters in word 
2. Rhyming group simply counted letters in word think of words that rhyme 
3. Adjective group simply counted letters in word think of adjective for that word 
4. Imagery group simply counted letters in word visualize words you read 
5. Intentional group  simply counted letters in word explicitly instructed to memorize words 
- One IV= learning group w/5 lvls, DV= how many words were recalled 















,  The greater the between groups differences the less likely it is that this difference is caused
by chance
 The greater the within group differences the more likely it is that any between group
differences are caused by chance
 Hence, we need to make a distinction between differences within and between each of the
groups:
- MSerror—i.e. MSwithin
- MSgroup—i.e. MSbetween

How to calculate ‘MS’ (mean square) for one-way ANOVA:
1. Compute total sum of squares (SS)
SStotal= ∑ (Xij- X..) 2
- SS total= (all differences between groups + individual’s deviations from average – all
differences) 2
- Compute difference between ppts individual score + overall group average
2. Determine sum of squares that can be attributed to individuals
SSERROR= SSTOTAL – SSGROUP
- Compute for each group the difference w/overall mean + take square of that value
- After doing this for all groups add all squared values together X amount of ppts in each
group
3. Transform ‘sum of squares’ into ‘mean square’ (MS) divide SS values by df associated
w/them—done in order to correct for when SS become larger when sample sizes + amount of
groups increases
- For SSgroup df= total no of groups (k)-1
- For SSerror  df= total ppts-1  i.e. k(n-1)
- Then calculate MSERROR and MSGROUP
- MSERROR= SSERROR ÷ df error
- MSGROUP= SSGROUP ÷ df group
4. Make comparison between MS error and MS group to see how large the differences are
relative to each other



Larger the F value, the smaller the P value
5. Look up the p-value linked w/ratio by inspecting F-distribution table to check if groups differ
significantly

Assumptions of ANOVA:
 Homogeneity of variance= i.e. homoscedasticity implies pop variances underlying groups
should be same distributions between groups are roughly same (SD are same + width of
distribution)
- Levene’s Test reveals if there are significant differences between variances of each condition
 Normality= i.e. scores are normally distributed
 Independence of error terms= i.e. ppts scores are not influenced/independent from each
other if scores are not independent from each other (e.g. having similar attitudes to each
other)—have to conduct BS design

1.5: Be able to perform an ANCOVA=
ANCOVA in SPSS (P + L + R) - Statistical control can be used when comparing groups (co-variance i.e. sm that relates to
DV that is difficult to experimentally manipulate)
- When conducting experiments, researcher can account for factors by:
- Manipulating them (e.g. comparing treatment vs. placebo)
- Experimentally controlling them (e.g. random assignment)
- However, sometimes this does not work; often for quasi-experimental + correlational designs
- Covariate= continuous variable we want to control for e.g. age, SES, IQ, clinical history
- By controlling for such variables, we can examine whether differences between groups
emerge after controlling for these variables