week 11-statistical tests Flashcards
Comparing tow group means in an RCT
- assume groups at pretest are baseline
- difference between groups at posttest is due to treatment
- unlikely to be result of chance
Comparing tow group means in an RCT: parametric assumptions
- groups have normal distributions
- groups have variances
- interval/ratio data is used
- compare the difference between groups with a t-test
Statistical erroe
- all sources of variability within a data set that cannot be explained by IV (not necessarily mistake)
- applies to the random differences that are not explained by treatment
- source of error is variance - all subjects in groups do not respond the same
T distribution
- based on properties of normal curve (95% within 2 SD from mean)
- determine z-score of area of curve above and below that value and probability of obtaining score
- significance of difference between the two groups judged by ratio
- t=difference between means/variability within groups
- H0 is true if error variance increased and t ratio decreased
- H0 is false if error variance decreases and t-ratio is increased
Independent Unpaired t-test
- compare means from 2 independent groups (between subjects design)
- each group composed of difference set of subjects
- t=(X1-X2)/(Sx1-X2) = difference between independent groups/pooled variability within groups
- compare calculated t-value with critical value
- probability that calculated t-value will be larger than critical value is 5% or less
- t-value larger than critical value is considered significant and reject the h0
Two-tailed test (nondirection HA)
- new splint: X
- Stnadar splint: X
calculated t-vlue - critical value
- calculated t-value
paired t-test
- subjects serve as own control (within subject design)
- each measurement has a matched value for each subject
- determin if these are significantly different from each other
- t=mean of difference scores/standard error of difference scores
use of multiple t-tes
- compare two means
- if > 2 means should not compare using multiple t-tests
- making more comparisons means you are more likely to commit type 1 error
- 5% chance of type 1 error with each test
- cummulative error of multiple test >5%
ANOVA basics
- purpose: compare 3+ groups
- like t-test based on comparison of distance between group means and error of variance of each group
- parametric assumptions:
1. normal distribtuion
2. homogeneity of variance
3. interval/ratio data
One-way NOVA
- one IV (3+ levels)
- h0: uA=uB=uC…
- HA: there will be a difference between at least tow of the groups
- Calculates f-statistic
- conclusions: reject H0 if there is a significant different but you do not know where it is
Multiple comparision test: one-way ANOVA
- if a significant difference is found you must carry out a post hoc multiple comparison test to locate differences
two-way ANOVA
- 2 IV (factor a and factor b)
- null hypothesis for each main effect of the factors
- if significant difference is found a mulitple comparisions for 2-way ANOVA test should be conducted
Chi-Square: nonparametric test
- not based on assumptions about population distribution
- use rank or frequency information to draw conclusions about difference between population
- ## use with categorical variables (nominal data)
Chi-square assumptions
- frequencies represent individual counts
- categories are exhaustive and mutally exclusive
chi-squared exquation
- summation of (O-E)^2/E
- statistic based on differences between observed frequencies and frequencies that would be expected if null hypothesis were true
two-groups parametric tests
- independent t-test: between subject design
- paried t-test: within subjects design
3 + groups parametric design
- ANOVA: 1 IV
- Multifactorial: ANOVA (2-way, 3-way etc) with 2+ IV
- if significance is found then multiple comparisons to locate difference
Categorical data: nonparametric
- chi-squared: compares observed frequency with expected frequency
correlation research
- analysis of association between variables
- nonexperimental, no IV, observe existing phenomena
- interval/ratio data
Correlation coefficient
- test statistic: r - describes strength and direction of relationship between 2 varibles
- values range from -1 to 1
positive value for correlation coefficient
- positive correlation
negative value for correlation coefficient
- negative correlation
assumptions for correlation
- subjects scores represent normally distributed population
- each subject contributes a score for X and Y
- X and Y are independent measures
- X values are observed not controlled
- relationship between X and Y is linear
Hypothesis testing in correlation
- significant correlation is unlikely to have occured by chance
- H0: population correlation = 0
- HA: population correlation doesnot = 0
Interpreting correlation coefficients: strength of relationship
- .00-.25=little correlation
- 0.26-0.49 = low correlation
- 0.5-0.69 = moderate correlation
- 0.7-0.89 = high correlation
- 0.9-1.00 very high correlation
Interpreting correlation coefficients: coefficient of determination
- square of correlation coefficient r
- percentage of variance shared by the 2 variables
- if r=0.76, r2=0.58 that means 58% of variability in one variable can be accounted for be the other variable
Statistical significance
- probability that correlation coefficient would have occurred by change if there was no correlation between variables
- even weak correlations can have statistical significance