Research and Statistics Flashcards
Nominal
Labels, mutually exclusive, exhaustive
Ex: male and female
Ordinal
Rank ordering, distance between rating not equal
Ex: 1st, 2nd, 3rd place in a race
Interval
Equal intervals between ratings, no true zero point
Ex: temperature
Ratio
Equal intervals, has true zero point
Ex: 10MWT
Reliability
Consistent and dependable- results can be reproduced under same conditions
Random errors limit reliability
Systematic errors limit validity
Data must be reliable before it can be considered valid
SEM
Repeated measures on the same instrument tend to be distributed around the “true” score
Large SEM = low reliability
Small SEM = high reliability
Ex: BP readings 120, 140, 160 vs. 102, 104, 106
Validity
The extent to which a test measures what it is purported to measure
A test must be reliable to be valid - although a highly reliable test may be invalid
Ex: bull’s eye with x’s
Construct Validity
How well a test measures the concept it was designed to measure
Ex: pain, intelligence, QoL
Content Validity
Assesses whether a test is representative of all aspects of the construct
Usually refers to surveys/questionnaires
Ex: QoL outcome measures
Criterion-related validity
Compares a test with other valid measures- gold standard
Concurrent- measures done at same time yield same results
Predictive- comparison btw the test and another measure administered in the future
Floor Effect
A measure’s lowest score does not capture patient’s level of ability
Ex: FGA has a floor effect for complete SCI
Ceiling Effect
A measure’s highest score is unable to assess a patient’s level of ability
Ex: FIST to young athletes post concussion
Minimal detectable change
The minimum amount of change in a patient’s score that ensures the change is not the result of measurement error
Minimal Clinically Important Difference
The smallest amount of change in outcomes that might be considered meaningful to patient/clinician
Sensitivity
True positive rate
How good a test is at determining who has the disease- test is positive for those who have the condition
A test with high sensitivity can be used to r/o the disease–
SnNout it out
SenSitivity = Screening
Specificity
True negative rate
A test that is good at finding those who do not have the condition
How often a test gives a negative result when the person does not have it
High specificity = can Spin it in
SpeCificity = Confirming
Positive Predictive Value
Percentage of people who are positive on the diagnostic test who have the condition
Ex 100% of people with vestibular disorder test (+), then high positive predictive value
Negative Predictive Value
Percentage of people who are negative on the diagnostic test who do not have the condition
Positive Likelihood Ratio
Indicates how many times more or less likely a positive test result will occur in someone with the condition than in someone without the condition
True positive rate compared to false positive rate
> 10: large, often conclusive likelihood disorder is present
5-10: moderate likelihood disorder is present
2-5: small likelihood disorder is present
1: normal (useless test)
Negative likelihood ratio
Indicates how many times more or less likely a negative test result will occur in someone without the condition than in someone with the condition
True negative rate compared to the false negative rate
1: neutral (useless test)
0.2-0.5: small decrease in likelihood of the disorder
0.1-0.2: moderate decrease in likelihood of disorder
<0.1: large, often conclusive decrease in likelihood of disorder
Mean
Sum all scores and divide by number of scores
Best used with interval/ratio data (unskewed)
Mode
The scores which are most frequently represented
Best used with nominal data
Median
Middle score for a set of data
Best used with ordinal data, skewed interval/ratio
Range
Difference between the highest value and the lowest value
No insight into distribution of scores
Percentiles/quartiles
Value below which a given percentage of scores fall
Ex: 70th percentile- value below which 70% of scores fall
25 percentile = 1st quartile (Q1), 50 percentile = 2nd quartile, 75 percentile = 3rd quartile
Standard deviation
Measure of the distribution of scores around the mean
Low SD: scores tend to be close to the mean
High SD: scores spread out over a wider range of values
Ex: 6MWT mean distance of 1,200ft, SD: 30ft vs. SD: 300ft
Intraclass correlation coeficient
Measure of reliability of ratings
Describes how strongly units in the same group resemble eachother
Ranges from 0-1 (low to high agreement)
Generally ICC>0.75 is good reliability
Interferential statistics
Techniques that allow us to make generalizations about the populations the samples represent.
Descriptive statistics
Summarizes a sample rather than the population the sample represents
Descriptive statistics do not allow us to make conclusions about the population
Null hypothesis
Assumes there is no relationship between X and y
Experimental hypothesis
Assumes there is a relationship between X and Y
P-value
The probability that your results are d/t chance and are actually real regardless of test type.
P= 0.10 means there is a 10% probability your results are due to chance (and thus may be incorrect)
P= <0.05 means there is a 5% probability…- most commonly used
P= <0.01 means there is a 1% probability…
Used with tests of significant diff, relationships and tests that predict
Parametric statistics
Relies on assumptions about the population
interval and ratio data
Parametric data
2 IND groups: unpaired t-test
2 related groups: paired t-test
=/>3 IND groups: One-way analysis of variance (ANOVA)
=/>3 related groups: one-way repeated measures (ANOVA)
Nonparametric statistics
Does not rely on assumptions about the population
nominal or ordinal scales
2 IND groups: Mann-Whitney U test
2 related groups: Wilcoxon signed rank test
=/> 3 IND groups: Kruskal-Wallis analysis of variance by ranks
=/>3 related groups: Friedman two-way analysis of variance by ranks
Significance testing
Determines if groups are significantly different from each other
r2
Measures the % of variation in the values of the dependent variable that can be explained by the variation in the independent variable
Ranges from 0-1 expressed in %
Ex: r2= 0.8419 means that 84.19% of the variance in Y can be explained by changes in X
Confidence interval
Range of numbers of which you expect the true difference to fall.
Typically 95%, can be 98 and 99%
Measure of precision, range
Ex: CI=95% gives you 95% confidence that the true difference btw groups will fall btw a and b
Type 1 error
False positive
Ex: fire alarm goes off when there is no fire
Type 2 error
False negative
Fire alarm fails to sound when there is a fire
More severe- condition is present but test is telling us its not there
Statistical power
Alpha= 5% chance of incurring a Type 1 error
Beta= 20% chance of incurring a Type 2 error
Power= 1-beta
Increasing power:
-increase sample size
-increase alpha (from p0.5 to p0.1)
-larger effect size
Case Control Study
Patients with condition are compared to people without
Often rely on patient recall and medical records
Always retrospective
Cannot prove cause and effect
Case Report/Series
Detailed description of a series or single case = selection bias
No statistical comparison, only course of care
Cannot prove cause/effect
Cohort Study
Pt’s with the condition who undergo specific treatment are compared to those who either do not have the condition or did not have treatment
Can be prospective or retrospective
Provides evidence for cause and effect
Randomized Controlled Trials
Randomly assigns patients to treatment group or comparative group
Control: no treatment or sham
Treatment: other treatment
Strong evidence for cause and effect
Chi Square Test
Tests for the frequency of distribution.
Uses nominal data
Is the distribution of those categories any different?
Ex: are there a different amount of men in group A as compared with women in group B
Effect size
How big is the difference between groups? How much better did my patient get?
Larger effect size = larger difference
Ex: Cohen’s D, Hedges’ G, odd ratio, Pearson’s r