Research Flashcards
Internal validity
How well a study is done - whether it avoids confounding variables
Factors that influence internal validity -
Baseline effect Low statistical power Testing effect History effect Instrumentation effect Stat regression to mean Maturation effect Experimental attrition Diffusion of tx Contamination Assignment of groups Selection bias Compensatory rivalry Compensatory equalization Ambiguity of direction of causal influence
Construct validity
How well can inferences be made from it
Validity of the measurement tool
Factors that influence construct validity
Multiple tx interferance Order effect Hawthorne effect Experimenter effect Novelty and disruption effect Pre or post test sensitization
IV vs. DV
IV = the thing that is changed or controlled DV = depends on the IV (usually a measure)
Ex - Measure mile run times in Jim while drunk - Ind and Dep =
Ind = alcohol Dep = mile time
Scales of measurement - Nominal
Identity or classification Yes/No M/F Race Type of arthritis
Scales of measurement - Ordinal
Rank without equal distance between items
Scale 0-10
Rank favorite foods
MMT
Scales of measurement - Interval
Equal distance between numbers, NO absolute zero
Temp
Scales of measurement - Ratio
Equal distance between numbers, YES absolute zero
Weight
Height
Quartiles - Q1, Q2, Q3
Q1 = 25th (below median) Q2 = 50th (median) Q3 = 75th (above median)
Variance
Variability for the average squared distance that scores deviate from their mean
(sum of squared SD/# of scores - 1)
Standard deviation
Square root of variance 1 SD = 68% 2 SD = 95% 3 SD = 99% Summarizes variability in a set of data
Negatively skewed
Mean is less than median and mode
Bigger hump to the R
Positively skewed
Mean is greater than median and mode
Bigger hump to the L
Null hypothesis
No difference between the control and the experimental groups
Alternative hypothesis (H1)
Contradicts the null
Can be non directional (2 tailed) or Unidirectional (1 tailed)
Level of significance is usually set at
5%
P < 0.05
Z score is what
Standard score
Z of -1.0 means that the raw score is 1 SD below the mean
Z score and p value
Inversely related
If the z score is high, indicates that the test statistic is outside 2 SDs - so need to reject null
If the z score is low, it is within 2 SDs so we retain the null
Beta =
Probability of type II error
Retain the null (fail to reject it) when it is false
Alpha =
Probability of type I error
Reject the null when the null is true
Relationship of Beta and Alpha
The smaller you make alpha, the bigger beta is
If p value is less than alpha - what should you do
reject the null (there is a stat sig diff)
p value is what
Probability of getting data as extreme as you did by chance alone (if null is true)
P value of less than 0.5 means that there was less than 5% chance that the observed result was a fluke
If p is less than 0.5 you should
Reject the null (significance)
If p is greater than 0.5 you should
Retain the null (no significance)
Power =
the probability of rejecting a false null
Power = 1 minus beta
Rather than z scores, in health research we are more likely to see what
t test which is sample variance instead of population variance
T distribution
Like normal distribution, but greater variability in the tails
df = n minus 1
If the t value exceeds the critical t value, then you need to reject the null
(p and t are inversely related)
Reliability can be quantified with
T test ANOVA Pearson r R (true score variance/total variance) ICC SEM - this is the most useful for clinicians
Reliability
Precision
Consistency
Repeatable
Confidence interval - ex 95% CI means
95% of the time your measure will fall in this range
Validity
Accuracy
Measure what it says it measures
Criterion validity
How does it compare to the gold standard
Construct validity
How well can inferences be made from it
Validity of the measurement tool
Internal vs. External validity
Internal - how well study was done
External - how well it can be generalized to other people and other situations
Case control study
Select a control group that represents base population and then select a case group that has disease
Retrospectively look back and see differences in exposure
Cross sectional study
Sample a random group and look for disease and exposure status right then and there (like random survey)
Collect data on exposure and disease at a single point in time
Prevalence
Proportion of population with a disease at a specific point in time - existing cases
Incidence
Number of new cases over a specific period
Relative risk ratio
Risk of an outcome in exposed group/risk of an outcome in a non exposed group
If RR greater than 1, exposure is a risk factor
Odds ratio
Odds of exposure for cases/Odds of exposure for control
Sensitivity
SnOUT
High sensitivity = can rule it out
True Positives
How well a test detects those with disease
Specificity
SpIN
High specificity = rule in
True negatives
How well a test detects those without disease
Positive predictive value
Proportion of people with pos test who actually have disease
TRUE POSITIVE
Negative predictive value
Proportion of people with neg test who do not have disease
TRUE NEGATIVE
Correlation (r) =
Describes the relationship between 2 levels of an IV
Does NOT indicate causation
r = -1.0 to 1.0
0 = no correlation
- means one IV increases while other decreases
+ means both increase or both decrease
Pearson r =
Correlation coefficient
Test to see if r is large enough that it is unlikely to have occurred by chance
Regression
Used to explain changes in DV
Uses the line of best fit
Predicts the Y (DV) from X (IV)
Regression (R^2)
Coefficient of determination
Portion of total variance on measure that can be explained by variance in another measure
If r = 0.50, r^2 = 0.25 SO…. we can say that 25% of the variance in Y is accounted for by X
T test is used to
Compare 2+ levels of 1 IV
Ex = IV gender (M and F) DV (BP)
Ind vs. Dep t test
Ind = variables don't depend on each other (M and F) Dep = they do (pre and post)
one tailed vs two tailed t test
Two = has no direction in hypothesis One = has direction
ANOVA is used to compare
2+ levels/means
Uses an F statistic
One way ANOVA
2+ levels of 1 IV on 1 DV (so like independent t test)
If F stat is greater than critical value = significant diff
Repeated measures ANOVA
Used when subjects are tested more than once
Extension of dependent t test
Two way ANOVA
Compares 2+ levels of 2 IV on 1 DV
Factorial = both IVs are ind
Mixed model = one IV is ind, other is dep
Repeated measures = both IVs are dep
Chi square =
Analyzes freq of responses that are nominal
“ranked” - will be ordinal or nominal data
Non parametric test because not normal distribution
ANCOVA is used when
1+ IVs
1 covariate
1 DV
Discriminant analysis
1 IV (2+ levels) 2+ DVs
MANOVA
2+ IVs
2+ DVs
Tests for patterns (vs. ANOVA is testing for effects on an ind variable)
MANCOVA
1+ IVs
1+ covariate
2+ DVs
Levels of evidence
Systematic reviews and Meta analyses RCTs Cohort studies Case control studies Cross sectional studies Case series Case reports Ideas, opinions
