Pop Med Lectures 1-6

Front 1

Quantitative data- Continuous

Back 1

Measurement- numerical values
- Temperature, height, weight, BP

Front 2

Quantitative- Discrete

Back 2

- Counts
- Levels

Front 3

Sample vs. Population

Back 3

Sample: attempted representation of a population, a subset
Population:
- Target population- group you are trying to represent with data
- Study population- where you pull your sample from

Front 4

Probability distributions

Back 4

- Shape of probabilities of a random variables taking on specific values
--normal, chi square, binomial
- Common distributions are defined by PARAMETERS

Front 5

Types of distributions

Back 5

- Symmetric
- Skewed (right, left)
- Discrete (bars)

Front 6

Outliers

Back 6

Can have a big effect on the mean!

Front 7

What are the five quartiles?

Back 7

Q1= 0%
Q2 = 25%
Q3 = 50%
Q4 = 75%
Q5 = 100%

Front 8

What is the min, median, max in quartiles?

Back 8

Q1 = min
Q3 = median
Q5 = max

Front 9

What is the range?

Back 9

Q5 - Q1

Front 10

What is the IQR (in terms of quartiles)

Back 10

IQR = Q4-Q2, could be preferred to the range
- middle 50% data

Front 11

Qualitative data- 3 types

Back 11

Nominal- no order to the categories
Ordinal- inherent order to categories
Dichotomous- everyone falls in one category or the other

Front 12

Example nominal data

Back 12

Zip code
Race
gender

Front 13

Example ordinal data

Back 13

Stage of disease
grade in school

- can be described by medians and modes

Front 14

Example of dichotomous data

Back 14

gender
unmarried/ married

Front 15

what is variance?

Back 15

- spread of data

Front 16

How do you calculate variance?

Back 16

- mean of the squared deviation of the variable from its own mean
- Standard deviation = variance
- Range (IQR) is another way to display variance

Front 17

What is two number summary

Back 17

- median + IQR
- mean + SD

Front 18

Coefficient of variation

Back 18

100 * (standard deviation/ mean)

Front 19

Covariance

Back 19

- how much they vary together

Front 20

What does high correlation and covariance indicate?

Back 20

- dependence between variables

Front 21

what are other names for a normal distribution

Back 21

- bell curve
- bell shaped
- Gaussian distribution

Front 22

Describing data with two numbers

Back 22

Normal distribution- Mean, SD
Right skewed- median, range
Not continuous- group percentages

Front 23

Box and Whisker plot

Back 23

Lower/Upper boundaries- quartiles
Dividing line- median
Tails show how far data extends from quartiles

Front 24

What is a parameter? How is it useful?

Back 24

- A parameter is a true value that summarizes a characteristic of a population
- Need to estimate parameters of a population using data

Front 25

What are statistics used for?

Back 25

- Statistics are used to estimate parameters

Front 26

Accurate and valid methods- bias?

Back 26

no bias

Front 27

Precision, Reliable, Repeatable describe...

Back 27

method with low variability

Front 28

What is a confidence interval?

Back 28

- Range of values that a parameter could take on
- with 95% confidence, the paramer is between LB and UB

Front 29

Concept behind the confidence interval-

Back 29

If you sampled the population 100 times, with the same data structure, 95% of the time, true value would be in the interval

Front 30

Standard deviation vs. standard error

Back 30

SD describes spread of data
SE describes estimated parameter, depends on sample size

Front 31

Simple linear regression

Back 31

Y = B0 + B1 * (X+E)

Front 32

What does B0 represent

Back 32

B0 is the intercept

Front 33

What does B1 represent

Back 33

B1 is the slope of X, the linear relationship that X has with Y

Front 34

What does epsilon represent?
What is the goal of modeling?

Back 34

Epsilon is the error term
- The goal in modeling is to minimize error
- Terms in the model should explain away the error
- The more significant variables you find to explain outcome, the smaller the error term will be

Front 35

What is the use of the linear regression?

Back 35

- Study impact of X on Y
- Forecast the value of Y for a given X
- Subtract effect of X from Y using this model to create Y* and Y* is now adjusted for X

Front 36

What to look for in a graph-
Estimate
S.E.
P Value

Back 36

Estimate- positive or negative, clinically large?
SE- large?
P Value- less than 0.05? If so, it is a significant predictor of Y. These are called parameter tests.

Front 37

P values significance in regression

Back 37

- General linear hypothesis of whether the model has ANY significant predictors
--Does the model explain any significant variability in the outcome?

Front 38

Interaction

Back 38

- Cannot interpret either effect alone (age and insurance), must interpret both through interaction
- B1 + B12 are related, and must be considered together

Front 39

ADJUSTED values

Back 39

subtract the effect of Xis from Y using the linear model to create Y*, and Y* is now ADJUSTED for Xi's

Front 40

Analysis of variance
With two Wi's- what do you call this

Back 40

- Categorical variables that may explain variability in Y- factors
- W1, W2
- Two factor ANOVA

Front 41

With single factor ANOVA, and two levels, what do we want to conclude?

Back 41

- Does Y differ, based on whether you are in group A or B?
- Is BMI ( continuous outcome Y) different for men and women (categorical explanatory variable, with two groups)

Front 42

What is an indicator variable

Back 42

- An indicator variable, I (X), can only take on two values, (0, 1)
- Where I(X) = 1 if X is true, and I(X) = 0 if X is not true

Front 43

Simple regression model
Y= B0 + B1 (I(W1=B) + E

Back 43

Mean for A: B0
Mean for B = B0+ B1
Because of indicator variable, all observations take on values 0 or 1
- 1 is group B, 0 is group A
- E are individual differences (errors) from the mean

Front 44

When modeling a continuous outcome variable, you may use the terms:

Back 44

- Regression for continuous predictors only
- ANOVA for grouping or categorical variables
- ANCOVA for continuous AND categorical variables in the model

Front 45

What does a chi square test show?

Back 45

If two variables are associated

Front 46

Define degrees of freedom given R= rows and C=columns

Back 46

Degrees of freedom = (R-1)(C-1)

Front 47

What is the degrees of freedom?

Back 47

- Chi Square distribution is defined by mean, (degrees of freedom)
- Under null hypothesis, test statistic should be (R-1)(C-1)

Front 48

Comments on Chi-Square, cell size in 2x2

Back 48

- Sample sizes in each cell must be large enough for the test to be accurate
- Each cell size must be higher than 5 for approximation to be close enough to chi-square distribution

Front 49

What are cross tabs? Frequency tables?

Back 49

- Categorical variables: birth year, gender, race, diagnosis, zip code, school yr
- Cross tabs= Group sizes

Front 50

What is stratification?

Back 50

- Separating the analysis based on categorical variables (like tabular data)

Front 51

Odds ratio and relative risk

Back 51

two statistics used and calculated from 2x2 tables

Front 52

What is risk?

Back 52

- Number of diseased compared to those at risk

Front 53

If risk > 1, <1, =1

Back 53

>1, higher likelihood of disease if exposed
<1, lower likelihood if you are exposed
=1, indicate that likelihood of disease does not depend on exposure

Front 54

Odds Ratio

Back 54

- Odds are a measure of likelihood of something happening, as compared to the likelihood that it does not

Front 55

Odds ratio >1, <1, =1

Back 55

>1, likelihood of disease if you are exposed is higher
<1, likelihood of disease if you are exposed is lower
=1, likelihood is same as if you were or were not exposed

Front 56

What is the purpose of Fisher's Exact test?

Back 56

- Tests null hypothesis of NO ASSOCIATION in a 2x2 table

Front 57

Output/ Result Reporting- PARAMETERS

Back 57

- not simply slope parameters anymore
- using a link function of natural log
- Monotone: increasing unction
- Positive and negative signs still mean the same thing (negative estimates indicate negative relationships)

Front 58

For RR> 1, Parameter >0

Back 58

- HIGHER likelihood of outcome with an increase

Front 59

For RR<1, Parameter<0

Back 59

- lower likelihood of outcome with an increase

Front 60

RR=1, Parameter estimate = 0

Back 60

- would indicate that likelihood of outcome does not depend on variable

Front 61

Confidence interval- look to see if 0 or 1 is in it

Back 61

- if 1 in interval for RR
- if 0 in interval for parameter estimate

Front 62

In tabular (categorical data) what are outcome possibilities

Back 62

- counts on tables
- measures on tables

Front 63

Parameters in these models- what are thy?

Back 63

- NOT slopes
- Transformed rate ratios that act multiplicatively on the OUTCOME