Ger 1000 Tut 1 Design of Studies

GER 1000 tut 1 Design of studies

Chapter summary:

Association.

• If rate (A|B) does not equal to rate (A|not B), the A is associated with B
• Positive Association. If rate (A|B) > rate (A|not B), then A is positively associated with B
• Negative association. If rate (A|B) < rate (A|not B), then A is negatively associated with B
• No Association. If rate (A|B) = Rate (A|Not B), then A is not associated with B.

Pick same things from diff bags/groups. Allows us to compare.

Causation and association. If A is a cause of B, then A must be associated with B.

At the same time, if A is associated with B then at least one of the following must be true:

1. A is a cause of B
2. B is a cause of A
3. Some C is a cause of both A&B

Association does not imply causation, but causation implies association.

*Note: Associations found in controlled experiments and observational studies maybe causations, but easier to deduce causation from controlled.

Symmetry rule.

1. If  Rate (A|B) ≠ Rate (A| Not B)

then Rate (B| A) ≠ Rate (B| Not A).

1. If Rate (A| B) > Rate (A | Not B),

then Rate (B| A) > Rate (B| Not A).

1. If Rate (A| B) < Rate (A| Not B),

then Rate (B| A) < Rate (B| Not A).

1. If Rate (A| B) = Rate (A| Not B),

then Rate (B| A) = Rate (B| Not A).

1. If A has no association to B, then B has no association with A

Sandwich Rule:

Sandwich Rule. If Sample S is sliced into n number of mutually-exclusive sub-samples, and r1 is less than or equal to r2, r2 is less than or equal to r3, ..., and rn-1 is less than or equal to rn, then R is sandwiched in between r1 and rn (i.e., r1 ≤ R ≤ rn).

In other words, the overall rate (R) is sandwiched in between the smallest sliced rate (r1) and the largest sliced rate (rn).

Notes. The term ri refers to the rate of Event A among sub-sample i (i.e., rate (A | si)); and the term R refers to the rate of Event A among Sample S (i.e., rate (A | S)).

Confounding.

Confounding. If Event C is associated with both Event A and Event B, then C is a confounder of any potential association between A and B. Usually “sex” and “age” are common confounding factors.

May affect our understanding of the exposure and response

• We usually want to ‘remove’ effects from confounders

• Must be associated with both exposure and response

[pic 1]

In order to control for C, researchers conduct slicing. Alternative methods include regression and sample- size adjustment. Also do randomized sampling with Double Blinding

Slicing (sample size adjustment) – When a researcher uses slicing to control for C, he or she separates the data into smaller and more homogeneous sub-groups based on different levels of C, and examines the association between A and B within each group separately.

Simpson’s paradox: A trend that appears in a group of data but reverses (or disappears)

when the group is broken down into sub-groups and viewed separately. Relationships between percentages in subgroups can also be reversed when the subgroups are combined.

Basically, it refers to a phenomenon in which the sign of an association between Event A & Event B changes among slicing Event C.

Experimental studies

In experimental studies, researchers assign participants to either a treatment group or a control group.

Type of studies (Controlled vs Observational)

Controlled Studies

Controlled Experiments

• investigator has power to regulate factors
• Investigator assign subjects into “Control Group” and “Treatment Group” for comparison
• Ethical problems may arise

• Who should be in treatment/ control group?

• Bias may arise

• What is the confounding factor?

Randomisation – If a researcher assigns participants to either a treatment group or a control group in a random manner, then he or she has conducted a randomised controlled experiment.

By randomly assigning subjects, if the number of eligible subjects is large, it is very likely that the groups are similar in all respects, hence confounding is minimized

Non-Randomisation – If the researcher assign participants in a non-random manner, then he or she has conducted a non-randomised controlled experiment.