Midterm Flashcards
F test
higher F, stronger relationship
tests whole model
T value
for a single predictor
higher t value, more likely alternative hypothesis
higher n, t distribution closer to normal distribution
Type II Error
false negative
beta
larger sample size, lower beta, greater power
lower alpha, higher type II error, lower power
Type I Error
false positive
alpha (significance level/risk)
lower alpha, higher type II error, lower power
Power of the Test
1 - beta (1 - false negative/type II error)
larger sample size, lower beta, greater power
lower alpha, higher type II error, lower power
Non-Parametric Tests
Rank Sum
Bootstrap
Chi-Square
Mann-Kendall
Spearman’s Rho
Kendall’s Tau
Theil-Sen line
Parametric
two-sample/paired t-test
Pearson’s r
Rank-Sum Test
non-parametric
distribution-free
invariant to monotonic transformation (log, sqrt etc)
Ho: Prob(x > y) = 0.5 (median(x) = median(y))
T - Test
assume normal distribution
comparing two groups (no paired samples)
Ho: same mean
Bootstrap
resampling w/ replacement
non-parametric
Ho: P(statX > statY) = 0.5
Mann-Kendall Test
monotonic trends
any distribution (but no ACF)
comparing sequential vals (S = pos. change - neg. change)
Ho: S ~ normal distribution
calc Z-score based on S if Z > Z(1-a) then reject null
Sen’s Slope Test
compare all pair-wise slopes
Ho: central (1-alpha) ~ normal distribution
QQ Plot Skew
u shape – positive/left skew
n shape – negative right skew
QQ Plot Kurtosis
lower left tail + higher right tail – pos. kurtosis (less outliers/ peakier)
higher left tail + lower right tail – neg. kurtosis (more outliers/ flatter)
Pearson’s R
linearity btwn 2 variables
assume consistent y variance
Ho: r = 0
t-r > t-crit from (n-2) dist – significant
Chi-Square Test
for categorical values
greater than crit than reject null
Spearman’s Rho
non-parametric
rank-based
for monotonic
Ho: rho = 0
p < alpha then reject null
Kendall’s Tau
rank-based
for monotonic
calc Z-score based on tau if |Z| > Zcrit (Z alpha/2)
Mutual Information
related to entropy
KL divergence
not limited to monotonic
Autocorrelation
serial correlation
