last minute revision Flashcards
What is experimental psychology?
The use of scientific methodology to measure individual responses in a controlled situation or experiment to investigate the mind and or behaviour.
What are the steps in the research process?
- Develop research question based on initial observations and generate theory and hypothesis.
- Identify IVs and DVs.
- Design study to collect data testing the theory (Between/within subjects).
- Analyse data and graph data or fit a model.
What makes a good statistical model?
Fitting the data well.
What is a population in research?
A group of individuals that we want to generalise findings or a statistical model to.
What is a sample?
A subset of the population which is studied to infer info about the larger population.
What are the steps to calculate standard deviation?
- Calculate the sum of deviations.
- Square each deviation.
- Divide by df.
- Square root these.
What does high SD mean?
High deviation…platykurtic.
What does low SD mean?
Low deviation…leptokurtic.
What is ratio data?
Absolute 0…no values below it.
What is interval data?
Continuous with no meaningful 0.
What is ordinal data?
Ranked data 1,2,3,4,5,6,7 etc.
What is nominal data?
Categorical.
What is mesokurtic?
Normal distribution - symmetrical, same median mean and mode.
What is negative skew?
Lower mean, higher mode.
What is positive skew?
Higher mean, lower mode.
What is platykurtic?
Thinner tails…lack outliers.
What is leptokurtic?
Thicker tails…too many outliers.
What is z-skew?
Skewness/std error skewness.
What is z-kurtosis?
Kurtosis/std error kurtosis.
What range do z-skew and z-kurtosis values have to fall in to be normally distributed?
Z-scores have to be within the +/-1.96 range.
What are the types of statistical tests?
Independent samples t-test and then paired samples t-test.
How to solve normality issues?
- Check for outliers and remove or manipulate them.
- Transform the data with a mathematical function.
- Use a non-parametric test.
Why do non-parametric tests not mind if data is not normally distributed?
Because they rank the data, they don’t make assumptions about distributions and are not affected by outliers.
What is variance?
An estimate of average variability (spread) of a set of data.
What are degrees of freedom?
Number of people who can choose - 1.
What are examples of normally distributed data?
Babies birthweight, height, IQ
What do standard deviations mean for a normal distribution?
+/-1 sd from the mean has 68% of data.
+/- 1.96 sd from mean has 95% of data.
+/-3 sd from mean has 99.7% of data.
What is a z-score?
(your score - mean score) / standard deviation.
Standard error
- Standard deviation of sample means.
- Tells you how widely spread sample means are around the population mean.
confidence intervals
use sample mean and standard error to estimate out the range of mean values we have 95% confidence that the true population mean lies
Large confidence interval
sample mean is further away from the true mean of the population
Small confidence interval
sample mean is very close to the true mean of the population
what do error bars on plots represent
- 95% confidence intervals
- or standard error of the mean
test statistic
variance explained by the model/variance not explained by the model
effect/error
how to calculate a t-test
difference in means/standard error
=
explained variance/unexplained variance
purpose of significance testing
to determine which hypothesis to reject/accept…null vs alternative
range p values can fall in
0-1
alpha level used in psych
0.05
type 1 error
we believe there is an effect when there isn’t
the probability is the alpha level 0.05
type 2 error
we believe there is no effect in the population when there is
the probability is the beta level usually 0.2
effect size
standardized measure of the size of an effect
allows people to objectively evaluate the size of an observed effect
measures of effect size
- for t-tests: Cohens d
- for Mann- whitney: r
-correlation: r - chi square: odds ratio
descriptive Stats
means, SD, etc
inferential Stats
t-test, anova, etc
Samples from same population
IV has no effect on DV, difference is due to chance fluctuations due to sampling
Samples from different populations
differences in dv are due to changes in iv
Why cant we run lots of t-tests?
at p<0.05 there is a 5% chance of making a type 1 error…if we ran 20 tests then one of them would be significant just through chance
Assumptions of an ANOVA
- homogeneity of variance (the variances are the same/similar) or sphericity
- data on an interval scale
- sampling distribution of means is normal
What measures homogeneity of variance in a between groups ANOVA
levenes
significant-assumption violated
non-significant-assumption met
What measures homogeneity of variance in a repeated measures ANOVA
Mauchly’s test of sphericity
significant-assumption violated
non-significant-assumption me
ANOVA
allows us to compare multiple conditions in a single, powerful test
explained variance/unexplained variance
explained variance in a between group anova
variance between the groups
unexplained variance in between-group anova
variance within each group…also known as error
how many sources of variance are there in a repeated measure vs between participants design?
repeated measure- 3
between participants- 2
sphericity
equality of variance of the differences between all combinations of related groups
(or levels)
How to know which correction to make on Mauchly’s sphericity test
if greenhouse Geisser is less than 0.75 we make a GG correction…if not you make a huynd-Feldt
planned comparisons
comparisons that are hypothesised before any data is collected
unplanned comparisons/post-hoc tests
pairwise comparisons that test every possibility
Bonferroni correction calculation
0.05/number of comparisons
which measure of error to use in ANOVA eta squared or partial eta squared
repeated measures anova partial eta squared
between groups anova eta squared
Remember this by saying…repeated measures is fewer people and so partial eta squared is used
cohens d effect sizes for small, medium and large
small effect= 0.01
medium effect =0.06
large effect =0.14
non-parametric equivalent for the one-way repeated measures ANOVA
Friedmans test
non-parametric equivalent for the one-way between groups ANOVA
kruskal-Wallis
ANOVA dfs
Between subjects:
df1= k (number of levels to the iv) -1
df2= n (total number of observations) - k
Within subjects/repeated measure:
df1= k (number of levels to the iv) -1
df2= (no of observations - 1) - (no of ppts -1) - (k - 1)
Kendals W
Friedman’s test effect size calculation
chi squared/
sample size x (no of conditions-1)
0.1 small
0.3 medium
0.5 large
What test is used for friedman post hoc testing
wilcoxon test with a bonferoni correction
what test is used for kruskal wallis post hoc testing
mann-whitney test
main effect
the effect of one IV on the DV
Difference between a one way and a two way anova
one way has one iv
two way has two ivs
mixed anova
one iv is within one iv is between
when to use mauchlys test
for within IV- only when 3 or more levels
how to transform a positive skew
square root the values
which rows can you compare in post hoc tests
1 + 6 or 2 + 5
What is statistical power?
- the probability that a study will detect an effect when there is one
- correctly reject the null hypothesis when it is false. i.e. probability of avoiding type 2 error (false negative)
how can statistical power be expressed
1- beta
low powered studies produce
- unreliable findings
- more false negatives
what contributed to the replication crisis
low powered studies
Winners curse
the statistical unreliability of positive results from small published studies…as statistical power increases, bias falls
how many published psych studies have been found to be reliable?
one third
replication crisis
when studies were repeated they didn’t yield the same findings
limitations of null hypothesis significance testing
- overreliance on p values
- p values aren’t standardised as they also reflect the sample size
- a small study may be non-significant solely because of the sample size
- encourages all or nothing thinking
power analysis/ priori power analysis
checks what sample size is needed to detect an effect of this magnitude at this probability level
why are underpowered studies problematic?
have a lower chance of correctly rejecting the null hypothesis. waste of research resources…ppts burdened…impact of results may not be ethical as it is based on unreliable research
how to calculate cohens d
group a mean- group b mean
/ pooled standard deviation
what impacts statistical power?
- effect size
- sample size
- acceptable error rates
open science
advocates for transparency, sharing, and inclusivity within scientific research
examples of open science
- open source
- open methodology
- open peer review
- open access
- open educational resources
- open data
what is an interaction
how the effect of one iv varied depending on the level of another iv
how to transform negatively skewed data
- reflect and then transform
- done by subtracting each value from a constant
- e.g. (largest value +1) - origional value
and then square root them all