17 Research and Statistics Flashcards by Amy Belevan

Hierarchy of evidence

1 Systematic reviewed and meta-analysis
2 RCTs
3 Cohort studies
4 Case-control studies
5 Cross-sectional surveys
6 Case reports
7 Expert opinion
8 Anecdotal

How well did you know this?

Not at all

Perfectly

Nominal measurement

Labels, mutually exclusive, exhaustive

Male and female

How well did you know this?

Not at all

Perfectly

Ordinal measurement

Rank ordering, distance between ratings not equal

1st, 2nd, 3rd place

How well did you know this?

Not at all

Perfectly

Interval measurement

Equal intervals between ratings, no true zero

Temperature

How well did you know this?

Not at all

Perfectly

Ratio measurement

Equal intervals, true zero

10-m walk time

How well did you know this?

Not at all

Perfectly

Reliability

Consistency, dependability
Random and systemic errors limit reliability
Data must be reliable before considered valid

How well did you know this?

Not at all

Perfectly

Standard error of measurement

How repeated measures on same instrument tend to be distributed around true score
Large SEM = low reliability

How well did you know this?

Not at all

Perfectly

Construct validity

How well test measures the abstract construct it’s supposed to measure, like pain, intelligence, QOL

How well did you know this?

Not at all

Perfectly

Content validity

How well content of test matches a content domain associated with the construct
Usually refers to surveys or questionnaires

How well did you know this?

Not at all

Perfectly

Face validity

Under content validity

Test appears to test what it’s supposed to

How well did you know this?

Not at all

Perfectly

Criterion-related validity

Compares test with other measures already validated (gold standard)
Concurrent- measured at same time as other
Predictive- compare to future measure

How well did you know this?

Not at all

Perfectly

Floor effect

A measure’s lowest score is unable to assess a patient’s level of ability

How well did you know this?

Not at all

Perfectly

Ceiling effect

A measure’s highest score is unable to assess a patient’s level of ability

How well did you know this?

Not at all

Perfectly

Normative data

Represents scores pulled from literature to provide normal values for specific variables within a population
Provides approximate guidelines

How well did you know this?

Not at all

Perfectly

Minimal detectable change

Minimum amount of change in a patient’s score that ensures the change is not the result of measurement error

How well did you know this?

Not at all

Perfectly

Minimal clinically important difference

Smallest amount of change in an outcome that might be considered important by patient or clinician

How well did you know this?

Not at all

Perfectly

Sensitivity

True positive rate
Likelihood that someone with condition will be positive on test
High = few false negatives

Specificity

True negative rate
Likelihood that someone without condition will be negative on diagnostic test
Low = more false positives

Positive predictive value

Percentage of people who are positive on the diagnostic test who have the condition

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 condition than without
True positive compared to false positive rate
Large >10
Moderate 5-10
Small 2-5
Neutral 1

Negative likelihood ratio

Indicates how many times more or less likely a negative test result will occur in someone without condition than in someone with
True negative compared to false negative
Large <0.1
Moderate 0.1-0.2
Small 0.2-0.5
Neutral 1

Measures of central tendency

Mean
Median
Mode

Best measures of central tendency for data

Nominal = mode
Ordinal = median
Interval/ratio (not skewed) = mean
Interval/ratio (skewed) = median

Percentiles/quartiles

25th percentile = Q1 50th = Q2 75th = Q3

Standard deviation on normal distribution/bell curve

68. 2% within 1 SD of mean 95. 4% within 2 99. 7% within 3

Intraclass correlation coefficient

Measure of reliability of ratings Describes how strongly units in the same group resemble each other Ranges 0-1 (low to high agreement) ICC >0.75 is good reliability Low if poor agreement between raters or if there is not much variability between subjects

Inferential statistics

Allow us to use samples to make generalizations about the populations the samples represent

Null vs experimental hypothesis

``` Null = No relationship between X and Y Experimental = Relationship between X and Y ```

p-values

Probability of observing your sample results given that the null hypothesis is true p = 0.05 means 5% chance of finding a difference as large as or larger than the one in your study given that null is true p<0.05 means reject null, X and Y are different

Parametric stats

Interval or ratio scales Assumes normal distribution Assumes variance in data for the samples compared are roughly equal

Nonparametric stats

Nominal or ordinal scales | Does not rely on assumptions about population

Significance testing for 2 independent groups

``` Parametric = unpaired t-test Nonparametric = Mann-Whitney U test ```

Significance testing for 2 related scores

``` Parametric = paired t-test Nonparametric = Wilcoxon signed-rank t-test ```

Significance testing for 3 or more independent groups

``` Parametric = One-way analysis of variance (ANOVA) Nonparametric = Kruskal-Wallis analysis of variance by ranks ```

Significance testing for 3 or more related scores

``` Parametric = One-way repeated measures ANOVA Nonparametric = Friedman two-way analysis of variance by ranks ```

r-squared

Measures percentage of variation in the values of the dependent variable that can be explained by the variation in the independent variable Ranges 0-1 r-squared = 0.84 means 84% of variance in Y can be explained by changes in X; remaining variance is random variability

Confidence interval

Range of values likely to encompass the true value Confidence level is probability that CI encompasses true value Higher CL is wider CI

Type 1 error

False positive | Fire alarm goes off without fire

Type 2 error

False negative | Fire alarm does not sound with fire

Statistical power

Probability of type 1 error is associated with level of significance (alpha) Probability of type 2 error is beta Power = 1-beta Power means x% chance that treatment effect will be detected

How to increase statistical power

Increasing alpha (p from 0.05 to 0.01) Increasing sample size Large effect size (can have smaller sample if expect large effects)