Statistical Tests Flashcards
What does Relative Risk/Risk Ratio do?
● Measure of risk based on comparison of disease (or other health outcome)
incidence in two distinct groups
● Ratio of probability (percentage) of event occurring (or not occurring) in exposed
group vs. non-exposed group
● Negative exposure (toxin)= incident rate exposed/unexposed
● Positive exposure (prenatal care)= incident unexposed/exposed
Compares subgroup vs. entire population
● Commonly used in RCT’s and cohort studies
● RR of 1 means there were not differences between groups
● RR <1 means risk of bad outcome decreased
● RR >1 means risk of bad outcome is increased w/ intervention
What are Odds Ratios?
Odds of disease between two groups (happening vs. not happening)
Estimate of association (estimates the RR)
Gambling
Compares odds of event in one group to odds in comparison group
Overestimates (this number is always higher than relative risk!!!)
Commonly used in case control studies in epidemiological research
What do both RR and OR Ratios do?
Both compare the likelihood of an event to occur between two distinct groups
How are RR and OR Ratios different?
RR is easier to interpret & is consistent with the general intuition; Comparison b/w subgroup & entire population.. rather than… subgroup & remained population
Case-Control Studies in context of RR Ratios:
Case-control designs: limit RR calculation due to cases being selected on basis
of disease rather than exposure (RR compares exposed to unexposed)
When is RR & OR Comparable?
When the studied is rare
When does OR overestimate RR?
OR overestimates RR when the disease is more common → should be avoided if
RR can be used
Hypothesis Testing:
What is Hypothesis?
Assumed proposition
Hypothesis Testing:
What is Alternative Hypothesis?
Alternative Hypothesis is a Prediction that some observed difference is significant & due to knowable cause
~Statement there is a difference between groups not attributed to chance alone
~CAN accept the alternative hypothesis!
Hypothesis Testing:
Null Hypothesis:
● Reject null hypothesis or fail to reject null hypothesis
YOU CANNOT ACCEPT A
NULL HYPOTHESIS (you can only find evidence against it)
● Prediction that observed difference is due to chance alone and not due to
systematic cause
● Statement of “no difference”
How to Test the Null:
Type I and II Errors (used to test Null)
Type I: False positive (alpha)
Reality: no difference between groups
Study shows difference: error!
Study shows no difference: correct!
Type II: False negative (beta)
Reality: difference exist between groups
Study shows difference: correct!
Study shows no difference: error
What is False Positive and False Negative?
False Positive = good quality item gets rejected
False Negative = poor quality item gets accepted
What is Nominal?
Nominal: label or category without rank or order (mutually exclusive)
Ex. Male/female/dead/alive/pass/fail
Uses chi-square test
Tests difference in proportions
What is Ordinal?
Ordinal: label or category with some meaningful order or sequence
Not measured- without definite boundaries/levels
Ex. Severity of disease
What is Linkert Scale?
Linkert scale: strongly agree, agree, disagree, & strongly disagree
Uses Mann/Whitney
Tests differences in rank order
What is Interval?
Interval: scaled measure with an arbitrary zero point (temperature)
Difference between levels is meaningful
T-test or ANOVA
Tests differences in means
Used for chemists i.e. so irrelevant (sorry chemists)
What is Ratio?
Ratio: scaled measure with an absolute/true zero point(test score)
Chi Square =
Nominal Data
Mann-Whitney =
Ordinal Data
ANOVA =
Interval/ratio with 3+ independent groups
T-Test/paired T-Test =
Interval/Ratio w/ 2 Independent Groups
Inferential Statistics:
….?
Confidence Intervals:
Tells you most likely range of unknown population average
Most likely range within which the true size of effect lays
Confidence that if anyone reproduced this study they would have same results
due to independent variable(s)
What 3 things Impact the width of CI (Confidence Interval)?
Confidence level: typically 95%
Variability: standard deviation
Sample size: smaller sample sizes generate wider intervals
P values do not affect
If the CI includes 1 the effect isn’t significant
What is P Value?
-Tests likelihood of differences occurring by chance alone
How is P level determined?
- P-level is predetermined probability researcher is willing to make a type I error
- AKA “Alpha level”
What does P < 0.5 mean?
-P < .05 means a 5% chance observed difference was due to chance & a 95%
confidence that results were due to independent variable(s)
What does P < 0.001 mean?
-P < .001 means a .1% chance observed difference was due to chance & a 99.9%
confidence that results were due to the independent variable(s)
What do we do with a P-Score equal to or less than 0.05?
P-SCORE EQUAL TO OR LESS THAN .05 MEANS WE REJECT NULL &
ACCEPT ALTERNATIVE
What are the effects of Size in Statistics?
● Statistical significance (p-value) indicates difference between two groups
● Effect size describes magnitude of difference
How much more effective?
● We can standardize difference and compare it to 0
● Standard measure that can be calculated from various statistical outputs
● Standardized mean effect, expresses the mean difference between two groups in
standard deviation units (Cohen’s d)
Need mean and standard deviation for each group
Values calculated for effect size range from -3 to 3 just like the standard
deviation
Standard interpretation: .8= large .5=moderate .2= small
What is Experimental Event Rate (EER)?
EER = event rate in the treated/affected group
What is Control Event Rate (CER)?
CER = event rate in the control/unaffected group
What does Absolute Risk Reduction (ARR) do?
ARR:
Compares treatment effectiveness (CER-EER)
Difference in outcome rates between control and experimental groups
Inverse of NNT
How much does the treatment reduce the risk?
What does Attributable Risk do?
measure of the prevalence of a condition or disease.
Given a group of people exposed to a risk, it’s the fraction who develop a disease or condition.
Put another way, AR is the cases that would be eliminated if the exposure were also eliminated.
Opposite of ARR(EER-CER)«
What is Relative Risk Reduction (RRR)?
RRR = Percentage of original risk removed; Reduced Risk (ARR/CER)
Not a good number/test (can inflate numbers or findings)
“More or less likely to happen”
Describe Number Needed to Treat-Prevent Bad Outcome:
Patients to treat for 1 to have a benefit
Number of patients needed to prevent one additional bad outcome
Ideal NNT=1 (this means everyone improves with treatment- higher is less
effective and negative # indicates harmful)
1/ARR (inverse of absolute risk)
CER =
control event rate: event rate in the control/unaffected group
EER =
experiment event rate: event rate in treated/affected group
AAR =
absolute risk reduction: compares treatment effectiveness (CER-EER) → How
much does the treatment reduce the risk
Sensitivity =
of people who have the disease and test +/
of people who have the disease
Highly sensitive tests catch the disease every time:
sometimes they are wrong (false +)
Good at detecting/screening
SnOUT =
(sensitive test if it is negative rules disease OUT)
Probability that subject with the disease will screen positive for that disease ?? what do they mean exactly????
Highly sensitive tests can catch the disease nearly every time-
rules OUT
disease when the result is negative
NOT dependent on populations
Sometimes they are wrong (false +): say you have the disease and you really
don’t
Good at detecting & screening
If you test negative you can be almost 100% sure you don’t have the disease
Specificity =
of people who do not have the disease and test -/
of people who do not have the disease
Highly specific tests are rarely wrong
But sometimes they miss the diagnosis (false -)
Good at being right/confirming the diagnosis
Spin (specific test if it is positive rules a disease IN)
Sometimes miss the diagnosis (false -): say you don’t have the disease and you really do (bummer)
Good at being right/confirming the diagnosis
*CANNOT HAVE A TEST THAT IS HIGH IN BOTH SENSITIVITY & SPECIFICITY;
INVERSELY RELATED
What is Positive Predictive Value?
Chance to have disease when test was positive
Chance that when test is + you actually have the disease
PV+: true positive/ (true positive + false positive)
What is Negative Predictive Value?
Chance to NOT have disease when test was negative
Chance that when test is – you actually do NOT have disease
PV-: true negatives/ (true negatives + false negatives)
Statistical Significance vs. Clinical Significance
Statistical- used in hypothesis testing
Clinical- practical importance of a treatment effect- whether it has a real, palpable,
noticeable effect ton daily life
Describe Critical Thinking Competency:
ASK: convert information needs into a searchable question
ACQUIRE: search with efficiency for best evidence to answer
APPRAISE: critically
appraise evidence
APPLY: results in clinical practice
ASSESS: evaluate clinical
decision making skills
