ANOVA Revision

Question 1

INTRO

Answer 1

• t-tests only compare 2 groups; +2 = NOT a t-test
• may be a particular pattern unidentifiable unless we use (ie) ANOVA

Question 2

WHAT DO ANOVA DO?

Answer 2

• 2+ groups
• discover whether central tendencies (averages) are reliably different
• detect overall effect
• offers tools for making special-purpose comparisons/trends aka. polynomial tests

Question 3

POLYNOMIAL CONTRASTS

Answer 3

QUADRATIC TRENDS
- smile shape
CUBIC TRENDS
- rolling hill
QUARTIC TRENDS
- wave
LINEAR TRENDS
- positive straight line

Question 4

QUADRATIC TREND

Answer 4

• (ie. Does recall fall w/delay but then recover?)
• sig trend provides evidence that above statement = true

Question 5

CUBIC TRENDS

Answer 5

• (ie. Does recall fall with delay but only once a critical point is reached, and then drops to a given, minimum level?)
• significant trend provides evidence that above statement = true

Question 6

2-WAY ANOVA

Answer 6

• 2 groups/lvls/conditions
• allows to judge if STATSIG
• unpacks individual effects of 2+ separate IVs to analyse data from more complex designs
• can also investigate differential effects of 2+ IVs by exploring the interaction effect
• aka. tells us whether influence of one variable on scores = modulated by changes in others (aka. if there are INTERACTION EFFECTS)

Question 7

INTERACTION EFFECTS

Answer 7

• often not straightforward; usually “it depends”
• may be a big effect under some conditions (ie. driving at night) but not in others (ie. driving in daytime)
• ANOVA main strength = uses interactions to provide qualified answers to these questions

Question 8

INTERPRETING MAIN EFFECTS

Answer 8

(DRIVING EG)
TIME ME
- mean day perf = reliably dif from mean night perf
- aka distance affected by driving in day/night
WEATHER ME
- mean clear perf = reliably dif from mean foggy perf
- aka. distance affected by driving in foggy/clear
TIME x WEATHER INTERACTION
- day/night dif in clear = reliably dif from foggy

Question 9

PLANNED VS UNPLANNED CONTRASTS

Answer 9

• if group ME = reliable then there is dif somewhere (BUT where?)
• CONTRAST TESTS determine this (planned/A priori or unplanned/post-hoc/A posteriori)

Question 10

PLANNED CONTRASTS

Answer 10

• designed before experimenter sees data
• aka. stat tests thus far

Question 11

UNPLANNED CONTRASTS

Answer 11

• devised after experimenter sees data
• reduce Type 1 error likelihood via increasing crit value of test stat
• aka. prevent conclusion of false effects
• ie. Newman/Keuls; Duncan’s Multiple Range; Tukey’s; Dunnetts’s/Scheffe