1/21
Looks like no tags are added yet.
T Test
compares 2 qualitative independent groups on 1 dependent quantitative measurement
ex→ male vs female test scores (2 categories, number set)
T Test vs Z Test
T → sample sizes are small, population SD is unknown, uses estimate of SE
Z → larger samples, known population SDs
Parametric T-Tests
carry an assumption of normality in their dataset
assume data forms a normal bell curve
only have a smaller sample, have to estimate the population’s spread
Types of T-Tests
one sample, independent sample, repeated measures
One Sample t-Test
comparing 1 group’s average against a known target/standard number
X fully know population values
1 group: 1 group, hypothesis, item
DF= n-1
ex→ comparing LDL levels in a study to recommended level, lily size to standard size
samples = interval/ratio data
Independent Samples t-Test
comparing 2 completely separate groups of people/things to e/o
scalar & independent of e/o
similar variance of samples
DF= n -2
ex→ diff b/w male & female dexterity, comparing SAT math scores of left-handed vs right-handed
Paired Samples t-Test
comparing the same group twice
can see how a group changes over time → pre & post
2 variables w/ values paired by subject
DF: (n/2)-1 OR pairs-1
ex→ drug improving condition, grades increasing after tutoring, introducing plants to inc biodiversity
Degrees of Freedom
number of data points that are free to vary once you know final avg
based on sample size
in a statistical calculation, typically calculated as the total number of observations minus the number of parameters estimated.
t-Test Output Formula
group 1 ([mean] ± [SD]) group 2 ([mean] ± [SD]) , t(df) = t, p value for test statistic
t Statistic
measures how big the difference is relative to the variation in the data
bigger number = bigger difference
P Value
prefer 2-tailed, tells probability that results happened purely by random luck (H0)
if p < 0.05, statistically significant/REAL (rejecting null hypothesis)
if p > 0.05, not statistically significant/X REAL (accepting null hypothesis)
Determining Which Row to Use
look at Levene’s test
sig > .05 → 2 groups have similar spread/= variances → TOP
sig < .05 → = variances X assumed → BOTTOM
Nonparametric Tests
X assume specific distrib for data, used when reqs for parametric tests X met
X need for normality
robust to presence of outliers
work fine on ranked/ordinal data
lower statistical power
calculating p-values is more complicated
Types of Nonparametric Tests
1-Sample Kolmogorov-Smirnov → tests normality of data
Mann-Whitney U (like independent sample)
Wilcoxon Rank-Sum (like paired sample)
Mann-Whitney U Test
A nonparametric test
sees if there’s a diff b/w 2 indep groups
assesses if the ranks of the 2 groups differ statistically
substitute for the independent samples t-test when assumptions X met
Reporting Non-parametric Tests
U (total) = U, p = exact sig
Which p-value should be used in a nonparametric test?
exact significance → calc from all potential distribs
Asymptotic Significance
calc w/ est curve, inaccurate for small but closer to exact w/ larger N/as it grows
small n = less precise
very large n = more precise
Monte-Carlo Significance
uses random process to estimate significance using areas under curve
more accurate than asymptotic
use if medium/large
When should a one-sample t-test be used?
comparing a sample to a hypothesized mean
When should an independent samples t-test be used?
comparing 2 groups on some value
When should a paired samples t-test be used?
comparing 2 variables w/in a set of subjects