Helo I'm desperate need of help this assignment it's due tomorrow, at the bottom I've included all the need information that's needed including the background of the assignment.m All your help is appreciated, thanks! Please just try to complete the assignment to your understanding, I have provided all the instructions and clarification of this assignment that I was given. Thank you!!
Background
The article Experiments That Changed Nutritional Thinking is a brief review of two areas of pioneering science in nutritional and metabolic physiology.
Good science gives me the excitements. Unfortunately much of the science in nutrition is cargo-cult science – it looks like science but the methodology is junk. This cargo-cult science has a common theme: measure the correlation of some behavior in humans and some health outcome and present as if the behaviorcausedthe outcome**. A good example is this paper(https://twitter.com/CNN/status/1118547206276096006){target="_blank"}.
The problem with the methodology ismissing confounders. There are two ways we can infer the casual relationship between two variables, say coffee consumption and probability of heart attack.
- Anobservational study, that measures
- some measure of coffee consumption over some period of time
- incidence of heart attack over the period of time
- some measure ofeveryvariable that is both correlated to coffee consumption and has a causal effect on heart attack.
With item 3, we can estimate thetrue effectof coffee consumption on the risk of heart attack. If we don’t have measures of every variable that meets the two conditions, then our estimate of the true effect is wrong (the technical term is biased). The wrongness may be small or large…there is no way to know. The problem with this method is that we will never, ever, ever measure all variables that meet conditions in item 3. Ever. So we will always be wrong using this method…sometimes a little wrong, sometimes a lot wrong, but we cannot know which. What most people in nutritional science don’t seem to recognize is thatsample size doesn’t matter. We could measure the association in elebenty billion people and the answer would still bejust as wrong(the technical way to say this is the estimate converges on the biased value instead of the true value). I wrote a slightly more mathy explanation of confounding here.
- An experiment withrandomized assignment to experimental treatment group. The beauty of an experiment is that we can estimate true effects without bias. For example, if we randomly assign 1000 people to each of the five treatment groups in the table then we can ignore all other causal factors because the expected correlation between assignment and every other variable in the universe is zero (this is the definition of random assignment).
Cups per day of brown water that tastes like coffee and coffee
brown water |
coffee |
|---|
4 |
0 |
3 |
1 |
2 |
2 |
1 |
3 |
0 |
4 |
When humans are the subjects of the experiments, the experiments are called Randomized Controlled Trials (RCTs). RCTs have problems that also create biases. For example – lack of compliance or dropouts. In an RCT, however, we have some control of these bias-introducing factors but in an observational design, we have no control because we are ignorant of all the confounding variables and all we need is one missing confounder to make our estimate really, really, wrong.
In this better know, you will pump your intuition about confounding using an Excel spreadsheet and following the prompts below.
Total points: 10
Learning goals: numeracy, critical thinking, understanding literature
The Spreadsheet
In this assignment we are simulating different worlds. In each world, there is some truth that you control. In all worlds, there are only two factors that potentially causally effect the probability of heart attacks: coffee consumption and exercise time.
Download the spreadsheet “Confounding.xlsx”. The spreadsheet contains a table with three cells in the column “Truth” that you change. These cells are the true effect of coffee on probability of a heart attack, the true effect of exercise on probability of heart attack, and the correlation between the amount of coffee drunk and amount of exercise. These values are highlighted with green. You change these values as directed below!
World 1 simulates a Randomized Control Trial. The other worlds all simulate observational studies. Think of these as a researcher measuring coffee consumption and heart attack incidence over some period of time but not measuring exercise because they are ignorant of the fact that exercise is a confounding variable. Since it is not measured, exercise is amissing confounderoromitted variable.
The assignment starts here:
For each world, write down the world number (eg. “world 2”) AND i. true coffee effect ii. true coffee conclusion iii. estimated effect (“study results”) iv. the study conclusion, and v. if the study conclusion is consistent with the truth, at least qualitatively.
- World 1. True coffee effect: 0.0, True exercise effect: -0.8, correlation = 0
This world simulates an RCT – in reality there will be some measurable correlation between the assignment and exercise but the bigger the sample the closer to zero this will be.
- World 2. True coffee effect: 0.0, True exercise effect: -0.8, correlation = -0.6.
This world simulates an observational study with the truth just like that in World 1.
- World 3. True coffee effect: 0.0, True exercise effect: -0.8, correlation = 0.6.
This world simulates an observational study with the truth just like that in World 1 except the correlation between coffee and exercise is positive instead of negative (this could happen with a different demographic of people).
- World 4. True coffee effect: -0.5, True exercise effect: -0.5, correlation = 0.6.
This world simulates an observational study where coffee has a protective effect.
- World 5. True coffee effect: -0.5, True exercise effect: -0.5, correlation = -0.6.
This world simulates an observational study where coffee has a protective effect as in World 4 but the correlation between coffee and exercise is negative instead of positive.