1/168
Looks like no tags are added yet.
Grounded theory (review)
Qualitative approach; broad, developed from the ground up (inductive); builds theory from data
IRB (review)
Institutional Review Board; protects research subjects
Confidence interval (review)
Range of plausible values (all possible effect sizes) for the population parameter
Power analysis (review)
Determines the sample size needed to detect an effect (power usually 0.80)
4 questions for choosing the correct statistical test
Q1: Compare groups or relate variables? Q2: More than one IV (comparison) / more than two variables (relational)? Q3: Type of data and statistical assumptions? Q4: Between or within subjects? (Q1 & 2 pick the chart; Q3 & 4 navigate it)
Comparison study
Compares two or more groups (cause-effect designs)
Relational study
Looks at how variables relate: associate, correlate, predict. Correlation does not equal causation
Basic comparison vs. complex comparison
Basic = only one IV. Complex = more than one IV (tests interaction effects)
Basic relational vs. complex relational
Basic = two variables correlated. Complex/predictive = more than two variables
Parametric tests
Stronger tests; require interval/ratio data AND met statistical assumptions
3 assumptions for parametric tests
1) Normality 2) Homogeneity of variance 3) Independence
Normality assumption
Data distribute like a bell-shaped curve
Homogeneity of variance assumption
Distribution of the data is roughly the same between groups (especially with multiple groups)
Independence assumption
Each person's response/data point is independent from all other participants
Non-parametric tests
Used for nominal/ordinal data OR when assumptions are violated (treat interval/ratio data as ordinal)
Comparison study: which variable(s) do you check assumptions on?
The DV only
Relational study: which variable(s) do you check data type/assumptions on?
ALL variables
Between-subjects design
One group vs. another group; compares different people who had different levels of the IV
Within-subjects design
Compares the same person against themselves; stronger, more robust, more sophisticated (e.g., pretest vs. posttest)
Longitudinal study
Within-subjects design; repeated measures over time with the same people
Matched pairs
Two different people matched on something related to the DV but treated as ONE person (one gets treatment, one control)
Crossover design
Combines between and within designs; robust
How many levels does the IV have? (examples)
2 levels = treatment + control OR 2 treatment groups. 3 diets = 3 levels. 3 diets + control = 4 levels
Interval/ratio DV, 2 levels, between groups
Independent samples t-test (compares means)
Interval/ratio DV, 2 levels, within subjects
Paired samples t-test
Interval/ratio DV, >2 levels, between groups
One-way ANOVA
Interval/ratio DV, >2 levels, within subjects
Repeated measures ANOVA or ANCOVA
Ordinal DV (or violated assumptions), 2 levels, between
Mann-Whitney (compares medians/ranks)
Ordinal DV, 2 levels, within
Wilcoxon or Sign test
Ordinal DV, >2 levels, between
Kruskal-Wallis
Ordinal DV, >2 levels, within
Friedman
Nominal DV, 2 levels, between
Chi-square or Fisher's Exact Test (compares counts)
Nominal DV, 2 levels, within
McNemar
Nominal DV, >2 levels, between
Chi-square
Nominal DV, >2 levels, within
Cochran Q test
Complex comparison: interval/ratio DV, all between groups
Factorial ANOVA or ANCOVA
Complex comparison: interval/ratio DV, all within subjects
Factorial ANOVA with repeated measures on all factors, or ANCOVA
Complex comparison: interval/ratio DV, mixed (between and within)
Factorial ANOVA with repeated measures on the last 1 or 2 factors, or ANCOVA
Complex comparison: ordinal DV
No tests available to analyze the data
Complex comparison: nominal DV
Only all-between design works: Log Linear. Within and mixed = none
Basic relational: both variables interval/ratio, no violated assumptions
Pearson's r or bivariate regression (scores)
Basic relational: both variables at least ordinal
Spearman or Kendall's Tau (ranks)
Basic relational: one or both variables nominal
Phi or Cramer's V (counts)
What counts as "at least ordinal" in basic relational?
2 ordinal variables, OR 1 ordinal + 1 interval/ratio that violates assumptions, OR 2 interval/ratio-looking variables that violate assumptions
Complex relational: normally distributed outcome, any predictors
Multiple regression
Complex relational: dichotomous outcome, all normally distributed predictors
Discriminant analysis
Complex relational: dichotomous outcome, some or all dichotomous predictors
Logistic regression
Continuous vs. dichotomous variable
Continuous = multiple variations/values. Dichotomous = one or the other outcome (two options only)
What do you need to know about the test-selection charts?
Don't memorize the charts; know the 4 questions, how to navigate the chart, and whether interval/ratio assumptions are violated
Purpose of statistical significance testing
Decide whether the change seen in the sample is due to the intervention or to randomly selecting an unusual sample ("Do we think something happened?")
Probability (in significance testing)
How likely the current result would be if the null hypothesis were true
Random sampling error
Natural deviations that occur when randomly sampling from a population; reduced by increasing sample size
Bias (sampling)
Flawed sampling procedure; researcher does not use a representative sample (skewed toward a specific subgroup)
Sampling distribution (CLT concept)
Theoretical distribution made by taking every possible random sample (n of at least 30), averaging each, and plotting the averages
Central Limit Theorem point 1
If sample size is sufficient, the sampling distribution tends to be normal regardless of the shape of the population distribution
Central Limit Theorem point 2
The mean of the sampling distribution equals the mean of the population distribution
Central Limit Theorem point 3
As sample size increases, the standard deviation of the sampling distribution (standard error) decreases
Standard error
SD of the sampling distribution; smaller SE = sample means more stable and clustered closer to the true population mean
68-95-99 rule
68% within +/- 1 SD of the mean; 95% within +/- 2 SD; 99% within +/- 3 SD
Null hypothesis vs. alternative hypothesis
Null = no effect. Alternative = shift right (increase) or left (decrease)
Effect size (on the sampling distribution graph)
Distance between the null and alternative means
Critical value
Cutoff for rejecting the null; depends on alpha level, one- vs. two-tailed, and the sampling distribution of the test statistic (z, t, chi-square, F)
Logic of rejecting the null
Is the result rare if the intervention didn't work? If yes, reject the null hypothesis
Type I error
Reject the null when it is TRUE (false positive); probability = alpha (level of significance), usually .05
Type II error
Fail to reject the null when it is FALSE (false negative); probability = beta, usually .20
Power
Ability of a test to correctly reject a false null; 1 - beta; usually .80
Alpha (alpha level)
Allowable probability of making a Type I error; usually .05 but can be .01 or .10
Correct decisions in the Type I/II error table
Reject a false null (power) and fail to reject a true null
Interpreting the p-value
Probability of getting our result if the null hypothesis is true
p-value less than or equal to alpha (e.g., p
Statistically significant; reject the null
p-value greater than alpha (e.g., p > .05)
Not statistically significant; fail to reject the null
How to read t(38) = 2.45, p < .001
t = test statistic used; 38 = degrees of freedom; 2.45 = calculated test statistic; p = probability of that statistic if null were true
Why is significance testing vulnerable?
A very large sample makes trivial effects statistically significant (limited clinical relevance); interpret alongside confidence intervals and effect sizes
Effect size
Strength/magnitude of an effect, relationship, or odds; answers "How big was it? How much does it matter?"
When to use raw mean scales for effect size
When the data are familiar, interpretable, and consistent across groups
Cohen's d vs. Hedges' g
Both = standardized mean difference between two groups. Cohen's d tends to overestimate the true effect; Hedges' g adjusts for bias in small samples
How is a standardized mean difference calculated?
Mean difference divided by a measure of variability (pooled standard deviation)
Why use Cohen's d / Hedges' g?
Shows how big the difference is relative to typical spread; compares results across studies with different scales
Cohen's d interpretation
Range about -2 to 2; 0 = no effect; positive = group 1 higher mean, negative = group 2 higher. Small 0.2, medium 0.5, large 0.8
Pearson's r
Strength and direction of the LINEAR relationship between two continuous variables
Point-biserial correlation (rpb)
Strength of relationship between one dichotomous and one continuous variable
r interpretation
Range -1 to 1. Small 0.1-0.3, medium 0.3-0.5, large > 0.5 (sign = direction)
Relative risk (RR)
Probability of outcome in the exposed group divided by probability in the non-exposed group; used in COHORT studies
Odds ratio (OR)
Odds of exposure in people with the disease vs. people without it; used in CASE-CONTROL studies
Interpreting RR and OR values
Range 0 to infinity; 1 = no effect; >1 = exposed at greater risk (positive relationship);
Confidence interval (CI)
Precision; a range of plausible values for the population parameter; shows how sure we are of the finding
Meaning of a 95% CI
If you repeated the study 100 times with different samples, about 95 of the CIs would contain the true population parameter
Typical CI levels
95% (alpha .05) or 99% (alpha .01); complementary to the alpha level
3 factors CI is calculated from
Sample size, response distribution, population size (calculated around the effect size)
How to write a CI
Confidence percentage, then lower and upper limits, e.g., 95% CI [5.32, 10.45]
CI contains the "no effect" value
Cannot be certain the effect is real (no effect = 0 for mean difference/correlation, 1 for OR/RR)
CI width and precision
Narrow CI = more precise; wide CI = less precise
What a CI tells you
If an effect occurred; the range of possible scores with other samples; whether the effect could be negative or positive; the precision of the estimate
Example: 95% CI [0.1, 0.9] for a standardized mean difference
Real effect (does not contain 0), only positive, but wide so not precise (effect could be very small or large)
Example: 99% CI [0.7, 3.5] for an odds ratio
No real effect (contains 1); exposure could be protective or risky; medium width
Ethnography
Researcher immerses in subjects' environment to observe/interact and understand their culture, behaviors, social interactions. Rooted in anthropology; insider perspective; cultural relativism
Ethnography example
Researcher living in a remote village for months to study social rituals, norms, and daily activities
Cultural relativism
Understanding behaviors, attitudes, and norms relative to a particular culture
Grounded theory (qualitative approach)
Systematic approach where the researcher constructs theory through methodical data gathering and analysis; patterns/themes emerge from data
Grounded theory key features
Codified procedures; inductive coding; codes then categories then themes