W4 - Stats Refresher Flashcards
What are the 2 most commonly used parametric tests in experimental work?
ANOVA
T-tests
Define variance
How spread out data is in relation to the mean + how close ind values are to the avg value.
What is variance expressed as?
Sigma squared
What is the variance of a data set describing?
Avg error between mean value + ind values of a data set
What is population variance describing?
Variance of the entire pop of interest
Usually hypothetical unless we can measure every person in the pop
How is population variance calculated?
Total deviance / Sample size
Define sample variance
Variance of our experimental sample
How is sample variance calculated?
Total deviance / degrees of freedom
Which is smaller than which?
Sample variance or population variance
Population variance is always smaller than sample variance.
What can be derived from variance?
SD
SEM
What is SD equal to?
Square root of the variance
Define SD
Variability of data around the mean
What % of the population fall within 1 SD of the mean?
~68%
What % of the population fall within 2 SD of the mean?
95%
How is SEM or SE calculated?
SD / Square root of Sample size
How might you examine variance in a visual manner?
By looking at error bars in graphs.
Define deviance
How different each value is from the mean
How is deviance calculated?
Calculate mean of data set
Each ind value - mean = deviance value
Square each deviance value
Add each squared deviance value to give the SS
Define the SS
Total dispersion or deviance of scores from the mean
Disadvantage to deviance
Size of deviance is dependent on how many values are in data set
Meaning its difficult to compare the SS from one data set to the SS of another data set with a different number of values.
Instead you would use avg dispersion rather than total dispersion. Avg dispersion is the variance.
SPSS output
What does it mean when the skewness value is more than double the SE of skewness?
MAY have a skewed distribution
SPSS output
What does it mean when the kurtosis value is more than double the SE of kurtosis?
May have a skewed distribution / non-normal distribution
What are the statistical tests of normal distribution?
Kolmogorov-Smirnov test
Shapiro-Wilk test (better for sample sizes <50)
What do statistical tests of normal distribution do?
Compare values in a sample to a normally distributed set of values with the same mean + SD.
Statistical tests of normal distribution…
p>0.05
Non-significant
Distribution of sample isn’t significant from a normal distribution so the distribution of data set is probably normal.
Statistical tests of normal distribution…
p<0.05
Significant
Distribution of data is sig different from a normal distribution so the distribution of data set is probably NOT normal.
What does it mean if tests indicate data is normally distributed?
Data meets the assumption of normality
What if the data is not normally distributed?
Check + remove for outliers ONLY if you have good reason to believe that person does not belong to the population you want to sample.
Use a non-parametric test (but means there will be a reduction in experimental power)
Easy way to normality –> TEST MORE PEOPLE
How can a data set be trimmed?
Prod a stem + leaf plot in SPSS to ID any outliers.
Most researchers use 1 or 2 rules to trim the data set.
What are the 2 rules most researchers use to trim the data set?
A % based rule
SD based rule
Trimming the data set
What does the % based rule involve?
Deleting a certain % of the values in your data set
i.e highest + lowest 10%
Mean of that is then the trimmed mean.
Trimming the data set
What does the SD based rule involve?
Calculate mean + SD
Remove values that are a certain number of SD from the mean
What is the purpose of t-TESTS?
To give a value for the sig of the differences between the groups or conditions
Independent samples t test
2 individual groups
2 experimental conditions + different participants assigned to each one
Assumptions to the independent t test
Data is continuous
Data is interval or ratio
Both groups drawn at random from pop = independent
Normally distributed data
Homogeneity of variance between the samples
Paired / 1 sample / Dependent t test
Comparing 1 group across 2 conditions
Assumptions to the paired / 1 sample / dependent t test
Data is continuous
Data is interval or ratio
Differences are normally distributed
Homogeneity of variance
Both groups drawn at random from pop = Independent
How is the output reported from SPSS for the independent samples t test
Each groups mean + SE/SD
Mean difference + CI range
T-value, df + p value
t(df) = t, p = significance value
How is the output reported from SPSS for the dependent/paired/1 samples t test
Each condition’s mean + SE/SD.
Outline difference + the CI range
T-value, df + p value.
What is the t score/ t value
Difference between groups : difference within groups
What does a larger t-score / t-value mean?
More difference between groups or conditions
What does a smaller t-score / t-value mean?
More similarity between groups or conditions
What would a t-value of 3 indicate?
Groups/conditions are 3 times as different from each other as they are within each other
When running a t-test what does a bigger t value indicate?
More likely the results are repeatable.
Define homogeneity of variance
Both groups have a similar spread of values around the mean
Levene’s test for homogeneity of variance
When is the value for the levene’s test NOT significant and what does this mean?
Not sig when p>0.05
So –> Accept assumption of homogeneity of variance
Use equal variances assumed from SPSS
Levene’s test for homogeneity of variance
When is the value for the levene’s test IS significant and what does this mean?
p<0.05
Assumption of homogeneity of variance is violated
Use equal variances not assumed
Give an example of a 1-tailed test
Do people perform better under drug A than drug B
Give an example of a 2-tailed test
Is there a difference in how drug A and drug B affect performance
In theory, which tailed test has how much more statistical power to detect an effect than the other?
In theory, 1-tailed tests have twice as much statistical power to detect an effect.
Which tailed test has more chance of finding a significant difference between groups or samples?
1-tailed tests
Why do 1-tailed tests have more chance of finding a significant different between groups or samples than 2-tailed?
Because you need less participants to reach significance.
In practise are 1-tailed tests used commonly?
No, rarely.
