Statistc Flashcards
1
Q
what are the levels of evidence
A
1 meta-analyses and Systematic review of RCTs
- randomize clinical control trials
- case study
- descriptive surveys
- expert opinion
2
Q
what is the difference between a systematic review and meta-analysis
A
- meta-analysis- pulls together all the data from the research and pools the data
- SR - pulls together all the data of expert opinion reviews.
3
Q
what are the keys to controlling bias in randomized control trails
A
- Random assignment test subjects
- specific manipulation of the intervention
- blinded assessment - outcome assessor is blinded
4
Q
what is the typical use of case-control study
A
- great for determining risk factors for a condition of interest
- retro anaylsis of group with condition of interest compared to a matched group without the condition of interest
5
Q
What are the two basic categories of statistical tests
A
- test of relationships - used to determine if there is a relationship between 2 or more variables
- tests of differences - used to determine if there is a difference between two or more variables
6
Q
what are the different types of research variables
A
- independent - variable research has control over (usually occurs prior to dependent variable)
- dependent - outcome of interest the research has little control over (object of study)
- extraneous - variable that can effect out come but are not independent
7
Q
What is the difference between internal and external validity
A
- internal - do the study outcomes reflect the relation ship between the independent variable on the dependent variable
- external - generalizability, can the test be repeated with different groups and achieve the same outcomes
8
Q
what is the null hypotheses
A
- there are no differences or no relationship between the variables or groupes tested
9
Q
what is the alternative hypotheses or research hypothesis
A
- there is a difference (either positive or negative) between the test variables or groups
10
Q
What are the types of error in research decisitons
A
type I (alhpa) - reject the null hypothesis when it is true (i.e. conclude there is a relationship when there isn't) type II - accept the null hypothesis when it is false
11
Q
how do you control for type one error rate
A
- set an alpha rate
- look at statistical significance (p-value and confidence intervals)
12
Q
what is alpha rate/level
A
- rate of type I error acceptance, typically 5%
- pre-selected threshold to detect statistical significance (probabilities of unknowingly rejecting the null hypothesis)
13
Q
what is p-value
A
- probability the study’s findings occurred due to chance
14
Q
what is the relationship between the p-value and alpha rate/level
A
the goal of the study is to achieve a p-value less than the pre-selected alpha rate so that you can conclude with a reasonable degree of certainly that you did not commit type I error
15
Q
What is the limitation of the p-value
A
- reduces findings/life to dichotomous “yes/no” conclusions
- the threshold is arbitrarily set and there is a big difference between .05 and .005 even if they both meet the alpha level
16
Q
what is a confidence interval
A
- range of scores that provides info about the statistic significance while characterizing the statistical perception (what ranges of score you can achieve to remain within the alpha rate)
- if the range includes 0 or negative numbers you cannot reject the null hypothesis
- the tighter the range the more precise the outcomes
17
Q
what is statistical power
A
- probability a statistical test will detect a relationship between 2 or more variables or differences between 2 or more groups
- probability you will achieve a type II error
- used to calculate the sample size
18
Q
What is the difference between descriptive statistics and inferential statistics
A
- descriptive - describes a population
- inferential - describes a sample and assumes normal normal distribution
19
Q
What are parametric statistics
A
a form of inferential statistics
- uses interval or ratio level data (can use ordinal data but there is no consistency between data points which is a problem for parametric statistics)
- typically focus on the mean
- key assumption is the data has a NORMAL distribution (but some statistical manipulation can account for this)
20
Q
What is the difference between interval and ratio data
A
interval - no zero point, no absence of the variable, temperature (there is no absence of temperature
ratio - zero point, time (there is a zero starting point)
21
Q
What are the characteristics of a normal distribution
A
- Symmetric distribution (nice bell curve) of data and center point of data is the mean
- The mean, medium (middle number of the scores), mode (most frequently occurring number) should be the same
22
Q
What does the standard deviation tell you about a normal sample
A
1 SD contains 68% of the observations
2 SD contains 95% of the observations
3 SD contains 99.7% of the observations
23
Q
What is the difference between a positively and negatively skewed distribution and how does this effect parametric statistics
A
Positive - most of the observational data falls above the peak of the curve distribution, income is an example the rich pull the high end a long way from the mode income
Negative - opposite
*parametric statistics don’t work well with this because the mean, medium and mode will not line up
24
Q
what is ordinal data
A
- the variables have natural, ordered categories and the distances between the categories is not known (5k finish times)
- The ordinal scale is distinguished from the nominal scale by having ordered categories. It also differs from interval and ratio scales by not having category widths that represent equal increments of the underlying attribute.
25
what are nonparametric statistics
- statistical tested used for data that does not have a normal distribution
- makes not assumptions and is more liberal in its analysis
26
What is the pearson product moment correlation (r)
- parametric test of relationship of a normal distribution of two variables
- assumes data is linearly related
- association DOES NOT equal causation (not a true measure of agreement)
- + r value positive relationship between the studied values
- r2 describes how much one values influences the other
- 1.0 perfect relationship
- 0 no relationship
- -1 perfect inverse relationship
27
what is the coefficient of determination
r2 value - pearson product moment correlation squared - used to explain the amount change one value with have in relationship to the second in a normal distraction
28
what is linear regression
- parametric test for a single independent variable and dependent variable
- predicts an outcome (variable 1) based on the value of a specific factor (variable 2)
- beta - the amount of change in variable 1 per unit of variable 2
- works best when the r value is close to one
29
what is multiple linear regression
- parametric test for multiple independent variables
- beta is same as linear regression and indicates the amount of change in dependent variable relative to independent variable
- requires a high r value to work best
- assumes in the independent variables are not correlated with each other
30
what is the spearman rank correlation coefficient
- non-parametric test similar to the Pearson r
| - used for ordinal data
31
what is logistic regression
- non-paramatsric test similar to linear region
| - used for nominal data or dichotomous data
32
what are the parametric assumptions
- normal distribution
- equal variance between groups
- independence of scores from one another
33
what is the independent t-test
- parametric test of differences
- used to determine if there are difference between variables
- usually a pretest
- can only compare to variables
34
what is a paired t-test
- parametric test of difference of two variables
- pre and post test design which subjects act at their own dependent variable (strength before and after intervention)
- assumes normal distribution
35
what is analysis of variance
- parametric test of difference for 2 or more groups
- assumes normal distribution
- only tells you difference between groups exists, but WILL NOT tell where the difference is (need post hoc analysis to do this )
- dichotomous answers
36
what are some examples of non-parametric test of difference and their parametric counter part
1. Mann-Whitney or wilson ran sum test - similar to t test
2. Wilcoxon signed rank test - similar to paired t test
3. Kruskal wallis test - similar to ANOVA
37
What is reliability
how reproducible are the results
38
what is a replication study
repeat the same study to determine reliability of results
39
what is the difference between intra- and inter- reliability
intra- within the same person
| inter- between people
40
What is the draw back of percent agreement reliability
- chance agreement
41
what is Cohen's kappa
- reliability statistic for nominal or categorical data
- determines proportion of agreement after removing chance agreement
- kappa= (proportion of observed agreement - chance)/ (1- chance)
- 1.0 perfect agreement
- 0 chance agreement
- negative less than change
42
how do you calculate chance agreement between two raters
- calculate marginal probabilities
- the sum of the proportions of rater 1 versus rater 2
- example
* rater 1 score 3 positive and 7 negative test for 30% positive and 70% negative
* rater 2 scores 5 positive and 5 negative test for 50% positive and 50% negative
* (.5*.3)+(.5*.7)= .15+.35 = .5 = 50%
43
what is margins probabilities
chance agreement - the sum of the proportions of rater 1 versus rater 2
- example
* rater 1 score 3 positive and 7 negative test for 30% positive and 70% negative
* rater 2 scores 5 positive and 5 negative test for 50% positive and 50% negative
* (.5*.3)+(.5*.7)= .15+.35 = .5 = 50%
44
what is the difference between observed agreement and chance agreement in reliability statistics
- observed agreement is the number of agree upon observations divided by the total number of observations
- chance agreement is the proportion of answers that could have occur by chance
45
how do you calculate kappa
* rater 1 score 3 positive and 7 negative test for 30% positive and 70% negative
* rater 2 scores 5 positive and 5 negative test for 50% positive and 50% negative
* observed agreement 3 yes/yes and 5 no/no for 8 of 10 trial agreement or 80%
* (.5*.3)+(.5*.7)= .15+.35 = .5 = 50% chance agreement
* (.8-.5)/(1-.5) = .3/.5 = .6
* 1.0 is perfect reliability, 0 is perfect chance
46
What is a weighted Kappa
Kappa calculation for greater than 2 categories
47
What is the major flaw of r (i.e pearson correlation coefficient)
does not distinguish between a case of perfect agreement and a case of systematic bias
- example: if the raters have a consistent disagreement (i.e. they always differ by a factor of 10) then the r value will look good
- it is not a reliability statistic
48
what is the intraclass correlation coefficient
- ICC is used for interval or ration data
- using an ANOVA model it is expressed as (varience with ratees)/(varience with ratees+error variance)
- 0 is no reliability and 1 is perfect reliability
49
what is a good ICC level
- greater than .75
| - poor less than .40
50
What is the difference between clinically significant and statistically significant
clinical meets the threshold for statistical progression
- clinically significant has true meaning to the patient
- large sample size can detect statistically significant changes that may not impact a patient in a clinically valuable way
