Lecture 6 Flashcards
classifications for probability distributions
- continuous or discrete
- univariate or multivariate
- central or non central
2 ways we can view probability distributions
- density function: can be seen by bell shaped curve
- probability function: can be seen by CUMULATIVE DISTRIBUTION FUNCTION, which in continuous function, involves density function –>
what does bell shaped curve indicate?
when we see a bell shaped curve, we are actually looking at the density function
why are f distribution and chi square distribution always negatively skewed?
because they can only have a NON NEGATIVE value on the x axis
how does a graph with positive skew look like
right tailed
‘facing’ right
most scores are on the left
how does a graph with negative skew look like
left tailed
‘facing’ left
most scores are on the right
what is mutually exclusive group?
independent groups
you can only belong to 1 group. each group is independent from another.
eg: you can only be in RMHI group or ARMP group.
what is mutually paired group?
dependent groups
each score in one group is LINKED to a score in the other group.
eg:
- twins = common dependency
- husband wife on marriage harmony
- different time points (with the same individual)
in terms of sample size, what differs mutually paired group to mutually exclusive group?
in mutually paired groups design,
the sample size in all the groups MUST be the same
in mutually exclusive group, you can have unequal sample size
assumptions for mean differences in 2 independent groups
- observations are independent
- obs scores are normally distributed
- variances in 2 groups are the same
what does balanced design mean
means the size of each group is the same
what is the diff between having balanced design and unbalanced design if the groups violate the homosdecasticity assumption
for balanced design, interpretation would still be robust if group variance is not consistent
if it’s unbalanced design, then interpretation would not be robust even if the homosdescaticity is only mild
“unstandardised confidence intervals are robust against mild to moderate non normality” true or false?
true
unstandardised confidence interval is ROBUST against these conditions
“standardised confidence are robust against mild to moderate non normality” true or false?
false.
standardised confidence interval is NOT ROBUST against these conditions
why is bonett functions useful?
it reports both observed and standardised mean differences
2 types of standardised mean differences
- hedges’ g (requires NORMALITY and HOMOGENEITY)
- bonett’s d (requires NORMALITY)
“we always assume homogeneity of variance unless p value for either test is small” true or false?
true.
if one test <0.05 and the other is >0.05 then we assume heterogeneity of variance.
what is the basis of analysis aka what are we calculating when groups are DEPENDENT?
we wanna analyse the DIFFERENCE between the scores (each involving both IV1 and IV2) for each individual
Assumptions for mean differences in 2 DEPENDENT groups
- independent observations
- normal distribution of observed scores
note:
HOMOGENEITY of variance is NOT relevant because the analysis is undertaken of the difference scores
why homogeneity of variance is not a relevant assumption in dependent groups?
because that’s the thing we are analysing
in what condition would the bonett test and hedges test result in the SAME standardised estimate value?
standardised estimate value in bonett and hedges test would be the same only if the correlation of the 2 sets of scores is 0.5
what is semipartial correlation
It is the correlation between observed scores on the dependent variable and that part of scores on an independent variable that is not accounted for by all other independent variables in the regression analysis.
Which of the following descriptions is closest in meaning to the concept of heteroscedasticity of residuals?
Non-constant residual variance for different PREDICTED values of the dependent variable. (NOT OBSERVED VALUES)
“we do not need to know how a null hypothesis test would be applied to a semipartial correlation” true or false?
true.
If we did know the standard error for the standardised regression coefficient and we undertook a null hypothesis test, then the obtained P value would be the same as that obtained for a null hypothesis test applied to a semipartial correlation. hence, the foregoing logically follows from knowing the relationship between (i) a semipartial correlation and a standardised regression coefficient and (ii) a confidence interval and a null hypothesis test.
