# Error, Misclassification, and Bias Flashcards

1
Q

What are the two types of error in epi inference?

A

Systematic bias and random error.

2
Q

What is systematic bias?

A

Distortion of results of a study in a CONSISTENT manner.

3
Q

What is random error?

A

Variability in data from one observation to the next. Dealt with via statistical methods.

Variance estimated in CI around estimate and presents random error.

4
Q

How can we define validity?

A

Relative absence of bias or systematic error. Estimates are more valid as validity goes up and bias does down.

5
Q

What is precision?

A

Define as relative lack of random error. More precise as sample size goes up.

6
Q

What are some combinations of precision and validity?

A

Good precision (close estimates) with good validity (closely around bullseye-which is the truth)

Good precision and poor validity (close estimates but not close to target)

Poor precision and good validity (far away from each other but close to target)

7
Q

What are the types of systematic bias?

A

3 types:

Confounding
Information Bias
Selection Bias

8
Q

What is information bias?

A

Misclassification or measurement error.

Recall bias, for example.

9
Q

What is selection bias?

A

Who is recruited. How they are recruited.

Who is retained and how.

10
Q

What is bias towards the null versus away from the null?

A

Bias towards the null: estimate for RR and OR is closer to 1 compared to the true value. Observed is smaller than true magnitude.

Bias away from null: estimate from RR and OR is away from 1. Estimate is LARGER than true value.

11
Q

What is sensitivity analysis?

A

Quantitative assessment to examine source of bias and amount of resulting uncertainty in studies.

12
Q

How do we assess systematic bias versus how do we assess random error?

A

Random error usually assessed through statistical analysis.

Systematic error assessed through sensitivity analysis.

13
Q

What is the general approach to sensitivity analysis?

A

4 steps:

• Hypothetical values used to create alternate estimates.
• Hypothetical values are given across a range to determine the range of possible inaccuracy around estimate.
• Range accounts for possible bias. What kind of estimate the readers should expect given a range around estimate we present as valid.
• CI estimates random error.
14
Q

How we account for confounding in design?

A

RCT, stratification, restriction.

15
Q

How do we account for confounding in analysis?

A

Standardization, stratification, and multivariate methods.

16
Q

What are the three main kinds of confounders?

A

Known and measured confounders, known and unmeasured confounders, and unknown confounders.

17
Q

What type of confounder does sensitivity analysis deal with?

A

Known but unmeasured confounders.

18
Q

What two things do we need for sensitivity analysis?

A

1) Estimated guesses for associations between outcomes and confounder among exposed and unexposed.
2) Prevalence of unmeasured confounder among people who are exposed versus unexposed.

19
Q

What does information bias refer to?

A

Results from measurement error.

20
Q

What are some types of information bias?

A
```Detection bias
Instrument bias
Recall bias
Social desirability bias
How we code it might be information bias```
21
Q

How can information bias occur?

A

How we collect data, how PPTs report data, and how we analyze data

22
Q

What is differential bias?

A

Differential misclassification occurs when the probability of being misclassified differs between groups in a study

23
Q

What is non-differential bias?

A

Non-differential misclassification occurs when the probability of individuals being misclassified is equal across all groups in the study.
Toward null ONLY.

24
Q

What is sensitivity?

A

The proportion of people with the disease who are classified as having disease:

TRUE positives/ALL positives (true pos + false neg)

25
Q

What is specificity?

A

The proportion of people without the disease classified as not having disease

True negs/Total negs (true negative +false positive)

26
Q

What do we need to address information bias?

A

Need sensitivity and specificity of exposure measurement among persons with outcome and also without outcome!