Stats Flashcards
Counterbalancing
Reduce Carryover effects
Latin Square
Practice/Testing Effects
Solomon Four-Group Design
Internal Validity
intervention causes changes in DV
Construct Validity
other factors in the intervention not intended to cause changes do
External Validity
interfere with generalizability
Reactivity
Hawthorne Effect
bx change b/c observed
External Validity
Rosenthal Effect
Construct Validity;
self-fulfilling prophecy
Statistical Conclusion Validity
low power (small sample size, poor interventions), unreliable measures, procedure variability, subject heterogeneity
Mean and SD Math
add or subtract from scores (Mean changes, SD does not)
multiple or divide (both change)
Z-Score
Z = (score - mean)/SD
Standard Error of the Mean
average amount of deviation between sample means
SDpop/square root of N
can be used to see if a treatment mean was far enough away to suggest it was the treatment that cause the difference and not just sampling error
Type I Error
reject null incorrectly
found a difference when there isn’t really one
Alpha = probability of Type I Error
direct with Power
indirect with Beta
Type II Error
accept null incorrectly
found no difference but there really is one
Beta = probability of Type II Error
Power
reject null correctly
found a difference and there is one
Power = 1 - Beta
- Large sample size
- Magnitude of intervention is large
- Random error is small
- Parametric Stats
- One-Tailed Test
F ratios
- 0 no significance
2. 0 + significant
Trend Analysis
analyzes non-linear data
obtained from quantitative IV
Nominal Data Tests
Test of Difference:
Chi Square
McNemar - correlated
Correlation/Association:
Phi
Ordinal Data Tests
Test of Difference:
1 Ind Gp - Kolmogorov
2 Ind Gp - Mann-Whitney, Median, K-S
3 Ind Gp - Kruskall Wallis
2 Cor Gp - Wilcoxon
3 Cor Gp - Friedman
Correlation/Association:
Spearman or Kendall rank correlation
Interval/Ratio Data Tests
Tests of Difference:
1 Ind Gp - t-test single sample
2 Ind Gp - t-test ind samples
3 Ind Gp - 1-way ANOVA
2 Cor Gp - t-test matched samples
3 Cor Gp - 1-way repeated measures ANOVA
2 IVs:
both Ind - 2-way or factorial ANOVA
1 Ind 1 Cor - mixed or split-plot ANOVA
2 Cor - repeated measures factorial ANOVA
Correlation/Association:
Pearson R
Coefficient of Determination
correlation coefficient squared
amount of variability in Y that is shared with, explained by, or accounted for by X
Assumptions of Bivariates
- Linear Relationship
- Homoscedasticity
- Unrestricted Range
Eta
correlation b/w X & Y is curvilinear
Correlation Tests
X and Y interval/ratio - Pearson r
X and Y ordinal - Spearman’s Rho or Kendall’s Tau
X and Y true dichotomy - Phi
X and Y artificial dichotomy - Tetrachoric
X i/r Y true dichotomy -Point-Biserial
X i/r Y artificial - Biserial
XY Curviliniar - Eta
Types of Correlations
Zero-Order - no 3rd
Partial/First Order - 3rd removed from both
Part/Semipartial - 3rd removed from one
Moderator Variable
affects the strength of the relationship
strength changes based on moderator
Mediator Variable
explains why there is a relationship;
when controlled for there is no relationship
Multivariate Tests
several predictors/IV (X) and 1+ criterions/DV (Y)
Multiple R
2+ Xs and 1 Y
Y is interval/ratio
1X is interval/ratio
Multiple Regression
predict Y based on Xs
best when low correlation b/w X’s and mod to high b/w each X and Y
compensatory technique
Multicollinearity
predictors (X’s) are highly correlated thus redundant
Stepwise vs. Hierarchiacal
Stepwise - all computer, fwd or bkwd (add/take away until change) - find fewest X’s
Hierarchical - person chooses adding order based on theory
Canonical R and Analysis
2+ of each variable
Analysis - prediction test
Discriminant Function Analysis
2x/1y but y is nominal
Loglinear (logit) Analysis
2x/1y all nominal
Correlational Causality
Path Analysis and Structural Equation Modeling (LISREL)
Tests of Structure
which variables in a set fit best together or form independent subtests
Factor Analysis
Cluster Analysis
Increase Reliability
- more items
- homogeneity of items
- unrestricted range
- less ability to guess
Reliability Types
- Test-Retest
- Parallel Forms
- Internal Consistency
Split-Half
Kuder-Richardson and Cronbach’s Coefficient Alpha - Interrater (Kappa)
Spearman-Brown Formula
used to predict how much more reliability if more questions (used for split-half)
Validity
Content - (content area)
Criterion - infer, predict, estimate outcome
Concurrent
Predictive
Construct - trait or construct (Convergent [monotrait, heteromethod] and Divergent/Discriminant [heterotrait, monomethod] validity)
Incremental Validity
amount of improvement when using a predictor test
best when mod base rate and low selection ratio