Statistics Flashcards Preview

Occupational Health > Statistics > Flashcards

Flashcards in Statistics Deck (69):
0

Study Power?

The power of a study is the probability of detecting a significant difference between treatments or study groups when there really is one.
Low power increases the likelihood of failing to identify a statistically significant difference when a real difference does exist.
High power (80% or more) is desirable .
Power is affected by sample size, etc.

1

Nominal data

In category, non-parametric

2

Ordinal data?

In order, with unequal interval,non-parametric

3

Interval data?

Equal interval
No absolute zero
Cannot compute ratio
parametric

Eg Tm in Celsius or Fahrenheit

4

Ratio data?

Equal interval
with absolute zero or true zero
Can calculate ratio
parametric

Eg. Wt, hight, Kelvin Tm

"NOIR"

5

Measurement of central tendency?

Mean
Median
Mode

6

Mean= Median = Mode, what distribution?

Normal distribution

7

Relationship of mean, median and mode in right (positive) distribution?

Right skewed -Tail on the right
Mean>Median>mode

(Rule of thumb: mean always follows the tail)

8

The relationship of mean, median and mode in left skewed distribution?

Tail is on the left of the distribution
Mean

9

For normal distribution, select statistic method?

Select Parametric statistics test
Eg. Student t-test, chi-square, ANOVA, ANCOVA, regression analysis

10

For non-normal distribution, eg. Bimodal, skewed, etc. test methods selection?

Non-parametric test eg.Fisher's exact test, McNemar test,Mann-Whitney U test, Wilcoxon's rank sum test, Kruskall-wallis test

11

Ways of obtaining random sample?

1. Simple random sampling
2. Systemic random sampling
3. Stratified random sampling
4. Cluster sampling

12

Bias?

Systemic error
Impacts internal validity

13

Chance

Radom

14

Confounder?

Associated with exposure (risk) and outcome
An independent risk factor for the outcome
Not in the causal pathway between the risk factor and disease

15

Power

The chance of finding an effect in your sample if it truly exist in the population.

Power is not a question in a study that shows a significant effects.

If a study results had failed to show a significant difference (p>0.05) between the two groups, one may wonder whether the study had sufficient power.

16

When apply to a population,
Given sensitivity and prevalence,
True positive =?
False negative =?

True Positive = Sensitivity x Prevalence
False negative = (1- Sensitivity) x Prevalence

17

When apply to a population, given Specificity and Prevalence,
True negative =?
False positive =?

True Negative = Specificity x (1- Prevalence)

False positive = (1- Specificity) x (1-Prevalence)

18

Regression toward the mean

In any group selected on a characteristic with substantial day-to-day variation, many will have values closer to the population mean when the measurement is repeated and worst pts will improve.

19

Baseline drift

Which occurs with measurement on certain machines that requires frequent calibration.

20

Hawthorne effect

A tendency among study subjects to change simply because they are being studied or watched.

21

1SD =? %
2SD =? %
3SD =? %

1 SD = 68% (Z score = 1)
2 SD = 95% (Z score = 2)
3 SD = 99% (Z score = 3)

22

When two events are independent, the probability of either will occur?

Is the sum of their probability, minus the probability that both will occur.
P (A or B) = P (A) + P (B) - P (A and B)

23

When two conditions are mutually exclusive, the probability that either one will occur is

The sum of their probability

24

Randomization

Assignment occurs by chance

25

ROC curve - Receiver-operator curve

X axis: 1 - specificity, or the false - positive rate

Y axis: Sensitivity

26

ROC curve is used to determine

Optimal Cut-off point for the respective test.
In general, the point closest to the upper-left corner, where sensitivity is highest and the false-positive rate is lowest, is chosen as the cut-off.

27

In ROC cure, the Area Under the Curve (AUC) is used to?

To calculate the diagnostic accuracy (best sensitivity and specificity) of the test, that is the probability of correctly identifying disease based on the result of the test.
The larger the area under the curve, the better the test.

28

Kappa statistic

Used for reliability studies, eg to assess inter-rater reliability or intra-eater reliability.
Used in assessing the degree to which two or more raters, examine the same data, agree when it comes to assigning the data to categories.

29

Effect modification

Occurs when one factor modifies the effect on outcome of another.

30

Confounder

Occurs when the association between two variables is distorted by the fact that both are associated with a third.
Eg. The association between coffee and lung cancer is distorted by smoking

31

CV (coefficient of variation)

CV = SD/X x 100%

1. Used for compare the relative spread of data for 2 variables (eg. Height and weight)
2. Used to evaluate precision of the measurement of a single variable (x-ray film reading by two physicians)

32

Histogram

For continuous variables

33

Bar graph

For categorical data

34

Scatter plot

For association

35

Types of random samples

Simple random
Systematic random
Stratified random
Cluster random

36

Simple random sampling

Every unit in the population had the same probability of being selected, chance alone determines whether a particular unit in the population is selected for the sample

37

Systematic random sampling

Every k th member is selected from the population

38

Stratified random sampling

- Population is divided into heterogeneous groups (strata) (eg. black, white, Hispanic, Asia) and a random sample is taken from within each group
- Ensures equal numbers of each strata in final sample.

39

Cluster random sampling

Population is divided into homogenous group (cluster) and a random sample of these groups is taken. eg a school, a community, etc

40

Z score

Z = (X - U)/sigma

Any normal distribution can be transformed to the standard normal to get a Z score for a given value X

41

Wilcoxon's signed rank test is an non-parametric equivalent of ?

Paired t-test

42

One sample t-test

To compare the sample mean with the mean of the population

43

Two samples t-test

To compare the mean of two groups

44

Paired t-test

To compare the mean of before and after

45

ANOVA

Used for more than two groups

46

Chi-square test

Compare two proportions

47

Fisher's exact test

Is used if expected count on a cell is less than 5

48

NcNemar's chi-square test

For paired proportions

49

Spearman's rank correlation coefficient is a non-parametric equivalent to ?

Pearson's correction coefficient

50

Coefficient of determination

% of variation in Y explained by X

51

Simple linear regression

Dependent variable is continuous
One independent variable

52

Multiple linear regression

Dependent variable is continuous
More than one independent variables

53

Logistic regression

Dependent variable is dichotomous
OR is used for estimation

54

Survival analysis

Time to the event
Hazard rate is use for estimation

55

Collinearity

Collinearity is a linear relationship between two explanatory variables.

Collinearity can result in unstable beta coefficient estimates.

56

Funnel plot

A graph designed to check for the existence of publication bias in systematic reviews and meta-analyses

57

When can Poisson distribution be used as a good approximation of a binomial distribution?

In general, p should be small , 15

58

Type 1 error
Or alpha

Reject H0 when it is true.

59

Type 2 error
Or beta

Accept H0 when it is actually false.

60

For Paired data (pre and post, paired), what test to choose?

For parametric data, using
- Paired t test ( pre and post, paired),

For non-parametric data, using
-Wilcoxon's signed rank test

61

To compare 2 group means, what test to choose?

For parametric data, using
- Student t test

For non-parametric data, using
-Wilcoxon's rank sum test (also termed Mann-Whitney U test.

62

To compare to proportions, what test to choose?

For parametric data, using
- Chi-square

For non-parametric data, using
- Fisher exact probability test
- used when at least 1 cell in a contingency table has an expected count s Chi-square test for paired proportion.

63

More than two groups, what test to choose?

For parametric data, using
- ANOVA

For non-parametric data, using
- Kruskal-Wallis test

64

For correlation, what test to choose?

For parametric data, using
- Pearson's correlation

For non-parametric data, using
- Spearman's correlation

Multiple regression
- more than one independent variable s

65

Time to event analysis

1. Kaplan-Meier analysis
2. Cox proportional Hazard Regression
- a combination of multiple logistic regression techniques with survival methods

66

Dependent variants categorical (binary, eg. Cured vs not cured), what test to choose?

Logistic regression

67

SD

How scattered the data is.

68

SEM

Precision of the mean.
How precise the data is.