Epi - Exam 3 Flashcards
What are the 3 statistical perspectives that can be taken by the researcher?
Statistical-Perspectives
- Superiority
- Noninferiority
- Equivalency
What are the 3 primary levels or groupings of variables (data)?
Variables - Primary Levels
- Levels:
- Nominal
- Ordinal
- Interval or Ratio
- ALL statistical tests are selected based on level of data being compared.
What are the 3 key attributes of data that define their level or grouping, and how are they assessed?
Variables - Key Attributes
- Attributes:
- Order/Magnitude
- Consistency of scale / equal distances
- Rational absolute zero
- Each attribute is assessed with a “Yes” or “No,” ‘does the variable have it?’
What is a nominal variable?
Variables - Nominal
- Dichotomous/binary; non-ranked; non-ordered; Named categories.
- No - order or magnitude.
- No - consistency of scale or equal distances (discrete).
- Nominal variables are simply labeled-variables without quantitative characteristics (or dichotomous/binary).
- ALL DICHOTOMOUS VARIABLES ARE NOMINAL.
- INCLUDES CATEGORICAL VARIABLES DESPITE NUMBER OF CATEGORIES.
What is an ordinal variable?
Variables - Ordinal
-
Ordered, rank-able categories, non-equal distance.
- Yes - order of magnitude.
- No - consistency of scale or equal distances (discrete).
- PAIN SCALE WILL ALWAYS BE ORDINAL ON EXAM.
What is an interval/ratio variable?
Variables - Interval/Ratio
- Order, magnitude, and equal distances (units).
- Interval = arbitrary zero value (0 doesn’t mean absence)
- Ratio = Absolute (rational) zero value (0 means absence of measurement value), i.e. physiological parameters).
- Yes - order or magnitude.
-
Yes - consistency of scale or equal distances (continuous).
- Living siblings (number) & personal age (in years).
How do the levels of data vary in specificity/detail?
Variables - Specificity/Detail of Levels
- After data is collected, we can appropriately go down in specificity/detail of data, but never up.
What are descriptive statistics?
Descriptive Statistics
- Non-comparative, simple description of various elements of the study’s data.
What are measures of central tendency.
Measures of Central Tendency
- Mode / Median / Mean
- Min / Max / Range
- Interquartile Range (IQR)
What is variance?
Measures of Central Tendency - Variance
What is standard deviation?
Measures of Central Tendency - Standard Deviation
What are the shapes of data distribution in graphical representations?
Shapes of Data Distribution
- Normally distributed
- Positively skewed
- Negatively skewed
Describe the shape of normally distributed data, in graphical representations.
Shapes of Data Distribution - Normal
- Symmetrical
- Mean/median are equal/near equal.
- Equal dispersion of curve (‘tails’) to both sides of mean.
What are parametric tests?
Stats Tests - Parametric
- Stats tests useful for normally-distributed data.
Describe the shape of positively skewed data, in graphical representations.
Shapes of Data Distribution - Positively Skewed
- Asymmetrical distribution with one ‘tail’ longer than the other.
- Mean > median = ‘positive skew’.
- “>” points right so tail points to right.
If the median differs from the mean, what does that mean for data distribution?
Shapes of Data Distribution
- The distribution is skewed.
- Mean > median = positive skew
- Mean < median = negative skew
Describe the shape of negatively skewed data, in graphical representations.
Shapes of Data Distribution - Negatively Skewed
- Asymmetrical distribution with one ‘tail’ longer than the other.
- Mean < median = negatively skewed.
- “<” points left so tail points to left.
What is skewness?
Shapes of Data Distribution - Skewness
- A measure of the asymmetry of a distribution.
- Perfectly-normal distribution = skewness of 0.
What is kurtosis?
Shapes of Data Distribution - Kurtosis
- A measure of the extent to which observations cluster around the mean.
- Kurtosis statistic:
- 0 = normal distribution.
- Positive = more cluster.
- Negative = less cluster.
What percentages coincide with standard deviation ranges?
Standard Deviation - Percentages
For proper selection of a parametric test, what are the required assumptions?
Parametric Test - Selection
- Required assumptions:
- Normally-distributed
- Equal variances
- Multiple tests available to assess for equal variances between groups.
- Levene’s test.
- Multiple tests available to assess for equal variances between groups.
- Randomly-derived & independent.
How is interval data, that is not normally-distributed, handled?
Interval Data - Not Normally-Distributed
- Use a statistical test that does not require the data to be normally-distributed (non-parametric tests), or
- Transform data to a standardized value (z-score or log transformation), hoping it will allow the data to be normally distributed.
- ALWAYS RUN DESCRIPTIVE STATISTICS & GRAPHS.
What is the null hypothesis?
Null Hypothesis (H0)
- A research perspective which states there will be no true difference between the groups being compared.
- Either accepted or not accepted, based on statistical analysis.
What is type 1 error?
Type 1 Error
- A.K.A. - alpha
- Not accepting H0 when it is actually true, and you should have accepted it!
What is type 2 error.
Type 2 Error
- A.K.A. - beta
- Accepting H0 when it is actually false, and you should have not accepted it!
In reference to statistical significance, what is power?
Power
- The statistical ability of a study to detect a true difference, IF one truly exists between comparison groups.
- The level of accuracy in correctly accepting / not accepting H0.
- Increases with a larger sample size.
How is sample size determined?
Sample Size
- Minimum difference between groups deemed significant:
- The smaller the difference necessary to be significant = the larger the N needed.
- Expected variation of measurement (known or estimated from past studies/population).
- Type 1 & type 2 error rates & confidence interval (usually 90%-99%)
- ADD IN ANTICIPATED DROP-OUTS OR LOSS TO FOLLOW-UPS.
What is a p value?
p Value
- Statistical tests determine possible error-rate or liklihood of chance in comparing difference or relationship between variables.
- A statistical test critical value is calculated.
- Test statistic value is compared to the appropriate table of probabilities for that test.
- A probability (p) value is obtained; based on the probability of observing, due to chance alone, a test statistic value as extreme or more extreme than actually observed if groups were similar (not different)
- The p-value is selected by investigators before the study starts (a priori).
Describe the statistical significance of the p value.
p Value - Statistical Significance
- If p value is lower than the pre-selected alpha value (customarily 5% (0.05)), then we say it is statistically significant.
- This means that the risk of experiencing a type 1 error is acceptably low if we reject H0.
How is a pre-set p value interpreted?
Pre-Set p Value - Interpretation
- The probability of:
- making a type 1 error if the H0 is rejected.
- erroneously claiming a difference between groups when one does not really exist.
- obtaining group differences as great or greater if the groups were actually the same/equal.
- obtaining a test statistic as high/higher if the groups were actually the same/equal.
How do you interpret and explain a p-value?
p Value - Interpretation and Explanation
- Explain it as you would a risk ratio with type of change and magnitude between groups, adding whether or not the data is statitstically significant or not statistically significant.
What is Levene’s test?
Levene’s Test for Equality of Variances
- Used to assess the equality or homogeneity of variances.
- Independent t-tests assume homogeneity of variances.
- Interpretation:
- >0.05 means variances are NOT significantly different, thus the variances are similar.
- 0.05 or less means variances are significantly different and thus there is no equality or homogeneity of variances.
What is a confidence interval?
Confidence Interval (CI)
- Most common selections are 90%, 95%, or 99%.
- Interpretation:
- We are 95% confident that the “true” difference or relationship between the groups is contained within the confidence interval range.
- Without p value:
- If CI crosses 1.0 (for ratios) or 0.0 (for absolute differences) = NOT significant.
- Calculated at an a priori percentage of confidence, including a high and low value, that statistically includes the real (yet unknown) difference or relationship being compared.
- Based on V/SD and N.
- Journals are moving away from solely reporting p values; or even showing them at all.
What are the 4 key questions to ask when selecting the correct statistical test?
Statistical Test Selection - Questions for Selection
- What data level is being recorded?
- Does data have order/magnitude?
- Does data have equal, consistent distances along the entire scale?
- What type of comparison/assessment is desired?
- Correlation -> correlation test
- Event-occurrence / time-to-event -> survival test
- Outcome prediction/association (OR) -> regression
- How many groups are being compared?
- Is the data independent or related (paired)?
- Data from the same (paired) or different groups (independent).
- Questions 2-4 get you around the other portions of each individual sheet.
What is a correlation test?
Statistical Tests - Correlation Test
- Correlation (r) - provides a quantitative measure of the strength & direction of a relationship between variables.
- Range from -1.0 - +1.0.
- -1.0 = strong negative correlation.
- 0.0 = no correlation.
- +1.0 = strong positive correlation.
- Partial correlation - correlation that controls for confounding variables.
- Types:
- Nominal correlation test = Contingency Coefficient.
- Ordinal correlation test = Spearman Correlation
- Interval correlation test = Pearson Correlation
- p>0.05 for a Pearson Correlation just means there is no linear correlation; may still be a non-linear correlations present.
- All correlations can be run as a “partial correlation” to control for confounding.
What is a survival test?
Statistical Tests - Survival Test
- “Changes over time.”
- Compares the proportion of events over time, or time-to-events, between groups.
- Commonly represented by a Kaplan-Meier curve (all types can be represented by a Kaplan-Meier curve).
- Types:
- Nominal survival test = Log-Rank test
- Ordinal survival test = Cox-Proportional Hazards test
- Interval survival test = Kaplan-Meier curve
What is a regression?
Statistical Test - Regression
- “Predict likelihood of some outcome” = regression test.
- Provide a measure of the relationship between variables by allowing the prediction about the dependent, or outcome, variable (DV) knowing the value/category of independent variables (IVs)
- Basically, helps to determine how well variables can predict outcome.
- Also able to calculate OR for a measure of association.
- Types:
- Nominal regression test = Logistic Regression
- Ordinal regression test = Multinominal Logistic Regression
- Interval regression test = Linear Regression
What is a partial correlation, and how is it used to control for confounding?
Partial Correlation
- Correlation that factors in the confounders, trying to keep them “mathematically quiet” to get a better idea of true correlation.
- Can be used to validate correlation test result by controlling for possible confounders.
- Example: Spearman’s returns small, but statistically significant, positive correlation. Partial returns even smaller postive correlation that is statistically not significant. Therefore initial correlation was not true.
What is the Spearman Correlation test?
Correlation - Spearman Correlation Test
- Correlation test for ordinal data.
What is the Contingency Coefficient test?
Correlation - Contingency Coefficient Test
- Correlation test for nominal data.
If you want to compare frequencies, counts, or proportions between the groups, what determines which tests are used?
Statistical Tests - Comparing Frequencies/Counts/Proportions
- Depend on 3 things:
- Type of data.
- Number of groups.
- Relationship:
- Independent data
- Paired/related data
- Pre- vs. Post, Before vs. After, Baseline vs. End
What statistical tests are used to compare independent, nominal data and what are they comparing?
Independent Nominal Data - Statistical Tests
- 2 groups:
- (Pearson’s) Chi-square test
- 3 groups:
- Chi-square test of independence
- Both tests compares group proportions and if they are different from that expected by chance.
- Assumptions-Both test:
- Usual chi-square (binomial; non-normal) distribution for nominal data.
- No cell count < 5 observations.
- For statistically significant findings (p<0.05) in 3 or more comparisons, one must perform subsequent analysis (post-hoc testing) to determine which groups are different:
- Multiple chi-square tests are NEVER acceptable, risk of Type I error increases with each additional test (almost guaranteed after 4-5).
- Chi-square test of independence
- 2 or more groups with expected cell count of < 5:
- Fisher’s Exact test
- Tests compare frequencies/counts/proportions.
What keywords indicate that data is paired or related?
Paired/Related - Keywords
- “Pre- vs. -post”, “Before vs. After”, Baseline vs. End”, etc.
If given both chi-square and Fisher’s Exact, what is the only situtation in which that can occur?
When the expected cell count is < 5. If you can confirm cell count < 5, then the chi-square is no longer valid and Fisher’s Exact is the right answer.
What statistical tests are used to compare paired/related nominal data and what are they comparing?
Paired/Related Nominal Data - Statistical Tests
- 2 groups of paired/related data:
- McNemar test
- 3 or more groups of paired/related data:
- Cochran
- Same as principle and assumptions as chi-square yet mathematically factors in concept of paired, or related, data.
- Bonferroni test of inequality (Bonferroni correction)
- Adjusts the p value for # of comparisons being made (very conservative).
- Cochran
- Tests compare frequencies/counts/proportions.
- “Pre- vs. -post”, “Before vs. After”, Baseline vs. End”, etc.
What statistical tests are used to compare independent, ordinal data and what do they compare?
Independent Ordinal Data - Statistical Tests
- 2 groups of independent data:
- Mann-Whitney test
- 3 or more groups of independent data:
- Kruskal-Wallis test
- The tests compare the median values between groups.
- Both also used for interval data not meeting parametric requirements.
- If 3+ group comparison significant, must perform a post-hoc test to determine where difference(s) is(are).
What statistical tests are used to compare paired/related, ordinal data and what do they compare?
Paired/Related Ordinal Data - Statistical Tests
- 2 groups of paired/related data:
- Wilcoxon Signed Rank test.
- 3 or more groups of paired/related data.
- Friedman test
- Tests compares the median values between groups.
- Each effective for non-normally distributed interval data or don’t meet all parametric requirements.
- If 3+ group comparison significant, must perform a post-hoc test to determine where differences are.
- “Pre- vs. -post”, “Before vs. After”, Baseline vs. End”, etc.
What are the post-hoc tests used for ordinal data?
Ordinal Data - Post-Hoc Tests
- Student-Newman-Keul test
- All groups MUST be equal in size.
- Compares all pairwise comparisons possible.
- Dunnett test
- Compares all pairwise comparisons against a single control.
- All groups MUST be equal in size.
- Dunn test
- Compares all pairwise comparisons possible.
- Useful when all groups are NOT of equal size.
What statistical tests are used to compare independent, interval data and what do they compare?
Interval Data - Statistical Tests
- Independent Data:
- 2 groups of independent data:
- Student t-test.
- 3 or more groups of independent data:
- Analysis of variance (ANOVA)
- Tests compare the means of all groups (along with intra- and inter-group variations) against a dependent variable.
- If 3+ group comparison significant, must perform a post-hoc test to determine where differences are.
- 2 groups of independent data:
- 3 or more groups of independent data w/ confounders:
- Analysis of Co-Variance (ANCOVA)
- Compares the means of all groups (along with intra- and inter-group variations) against a dependent variable while also controlling for the co-variance of confounders.
- Analysis of Co-Variance (ANCOVA)
What statistical tests are used to compare paired/related, interval data and what do they compare?
Paired/Related Interval Data - Statistical Tests
- 2 groups of paired/related data:
- Paired t-test.
- Compares the mean values between groups that are related.
- Paired t-test.
- 3 or more groups of paired/related data:
- Repeated Measures ANCOVA (1 DV)
- Compares the means of all groups (along with intra- and inter-group variations) of related data against a dependent variable.
- If 3+ group comparison significant, must perform a post-hoc test to determine where differences are.
- Compares the means of all groups (along with intra- and inter-group variations) of related data against a dependent variable.
- Repeated Measures ANCOVA (1 DV)
- 3 or more groups of paired/related data w/ confounders:
- Repeated Measures ANOVA:
- Compares the means of all groups (along with intra- and inter-group variations) against a dependent variable while also controlling for the co-variance of confounders.
- Repeated Measures ANOVA:
What are the post-hoc tests used for interval data?
Interval Data - Post-Hoc Tests
- Student-Newman-Keul test
- All groups MUST be equal in size.
- Compares all pairwise comparisons possible.
- Dunnett test
- Compares all pairwise comparisons against a single control.
- All groups MUST be equal in size.
- Dunn test
- Compares all pairwise comparisons possible.
- Useful when all groups are NOT of equal size.
- Tukey or Scheffe tests
- Compares all pairwise comparisons possible
- All groups must be equal in size.
- Tukey test slightly more conservative than the Stu.N.K.
- Scheffe test less affected by violations in normality and homogeneity of variances - most conservative.
- Bonferroni correction
- Adjusts the p value for # of comparisons being made (very conservative).
What is the Kappa test?
Kappa Tests
- Kappa statistic - a correlation test showing relationship or agreement between evaluators (consistency of “decisions”, “determinations”).
- Kappa interpretation:
- Kappa (K) value can be + or -; + = good agreement; - = poor agreement.
- +1 = the observers perfectly “classify” everyone exactly the same way.
- 0 = There is no relationship at all between the observers’ “classifications,”above the agreement that would be expected by chance.
- -1 = The observers “classify” everyone exactly the opposite of each other.
What is another word for correlation?
Relationship
How is research tracked before completion/publication?
Research Tracking
- Must be regsitered on clinicaltrials.gov and is assigned a NCT#.
- Reduces publication bias by helping to prevent failure to print due to nonfavorable outcomes.
What is needed to accurately assess a study?
Study Assessment
- To accurately assess a study, readers need complete, clear, and transparent information on the study’s methodology and findings.
- Historically, unfortunately, adequate and thorough assessments (by readers) of published studies can be challenging because authors may neglect to provide lucid and complete descriptions of that critical, necessary information.
Describe the CONSORT checklist.
Checklists - Interventional Studies
- CONSORT (Consolidated Standards Of Reporting Trials) - randomized interventional trials.
- May have extension documents for:
- Non-inferiority trials.
- Equivalence trials.
- Cluster trials.
- Pragmatic trials.
What are the checklists for analyzing studies?
Study Analyses - Checklist
- CONSORT (Consolidated Standards Of Reporting Trials)
- Randomized interventional trials.
- PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses)
- Systematic reviews of multiple randomized trials.
- STROBE (Strengthening the Reporting of Observational Studies in Epidemiology)
- Observational studies (cohort, case-control, cross-sectional)
- TREND (Transparent Reporting of Evaluations with Non-randomized Designs)
- Reporting evaluations with non-randomized designs of behavioral and public health interventions.
- REMARK (REporting Recommendations for Tumor MARKer Prognostic Studies)
- Tumor marker prognostic studies.
- GRIPS (Genetic RIsk Prediction Studies)
- Genetic risk prediction studies.
- STARD (STAndards for the Reporting of Diagnostic Accuracy Studies)
- Diagnostic studies.
- QUADAS-2 (QUality Assessment of Studies of Diagnostic Accuracy in Systematic Reviews (2nd ed.))
- Systematic reviews of multiple diagnostic studies.
Describe the STROBE checklist.
Study Analysis - STROBE Checklist
- STROBE (Strengthening the Reporting of Observational Studies in Epidemiology)
- For observational studies (cohort, case-control, cross-sectional).
- May have extension documents:
- Molecular epidemiology studies (STROBE-ME).
- Genetic association studies (STREGA)
- STrengthening the REporting of Genetic Association studies.
Fill in:
Fill in:
Fill in:
