A.2 Answer: TW 2 tutorial

Answers for Sect. 2.2

The study is observational, but because the researchers cannot determine the C (whether the person is a smoker or not). The critical element here is C, not O.
This is a mix of both C ('smokers and non-smokers') and O ('the median serum cholesterol').
External validity only refers to whether the sample represents the given target population, which is Australians. Whether the results apply for the entire world is irrelevant.
"Serum cholesterol" is not a variable; nothing here is varying. "Serum cholesterol" is just a type of cholesterol.
What actually varies--and so is the variable--is "the serum cholesterol concentration", or the "value of serum cholesterol".
This is not an experiment, since the individuals cannot be directed into the comparison groups (between smokers and non-smokers) by the researchers.
In the data file, each row is a unit of analysis and each column is a variable. So there will be two variables but not those listed: one column will record the smoking status (Yes/No) and one column will record the serum cholesterol concentration.
A confounding variable has to be related to both the response and explanatory variables.
The observer effect is about how the researchers might respond, not the individuals under study.

Answers for Sect. 2.3

The answers are, in alphabetical order: beach; blind; control; ethical; experimental; fire; help; nominal; randomly; relational; sample.

Answers for Sect. 2.4

Outcome: mortality rate or similar; Response variable: whether or not an individual baby survived.
Comparison: Between home and hospital births; Explanatory variable: Where the baby is born (this is what varies)
Observational (cross-sectional/non-directional) study.
Some are up for debate... The point here is that confounding variables are (potentially) related to both the response and explanatory variables.
1. The maximum temperature on the day of giving birth: Neither? Possibly confounding?
2. The health of the mother; Confounding (and hence extraneous, according to our definitions).
3. The distance to the nearest hospital: Confounding (and hence extraneous, according to our definitions).
4. The number of previous births by the mother: Not sure. Possibly confounding. I'd have to think more.
5. Baby's gender: Probably neither. Related to mortality (male babies have higher infant mortality) but not to the place of birth.
Not experimental.
Observational (cross-sectional/non-directional) study.
Possible RQ: "Among American mothers' births (P), is the neo-mortality rate (O) the same when giving birth at home compared to giving birth in hospital (C)?"
Cause-and-effect not reasonable (observational study).
Voluntary response. Data volunteered are likely to be more favourable than the data that was not volunteered. Limitations: many. For example, mothers who have been told to expect a difficult birth would probably opt for an in-hospital birth.
This report doesn't suggest it is safer.
The headline is accurate, but is certainly not the complete story as it implies cause-and-effect.

Answers for Sect. 2.6

Yes: "They were randomly allocated to take palmolein ("B9") or canola ("T4") crisps for the first 3 weeks, then (without a washout period) changed over to the other type, canola or palmolein for another 2 weeks".
It has: "the type of oil was known only by the food scientist..."
"the type of oil was known only by the food scientist..."
Probably.

Answers for Sect. 2.7.1

Use a multi-stage sample: Select a carpark at random, then a row of cars at random, then cars at random.

Answers for Sect. 2.7.2

The 'who' or 'what' which are observed, and for which data are collected: F. Units of observation.
A study where the researcher creates differences in the explanatory variable and measures the change in the response variable: G. Experimental study.
The result or effects of interest, across the population: B. Outcome.
What we 'do' to the individuals in the study: H. Intervention.
The question of interest to be answered by the study: D. Research question.
The larger group of individuals that are the focus of the study: E. Population.
The smallest independent 'who' or 'what' about which generalisations and conclusions are made, and for which information is analysed: C. Units of analysis.
A study where the researcher observes difference in the explanatory variable, and notices if these are related to changes in the response: A. Observational study.

Answers for Sect. 2.7.3

Variables: risk of dying from heart disease (or mortality rate? It's a bit ambiguous) as response; whether they get fish oil and/or Vitamin E as explanatory.
True experimental, as fish oil and Vitamin E are given to subjects, and the groups are determined by the researchers.
Randomization is not mentioned but probably used (a reporting issue). Control used (there is a placebo). Blinding: not stated. Blocking: none indicated. Study seems well done, so no obvious lurking variables (however, all subjects did change their diet). Group allocated by researchers, so a true experiment.
Since experiment, lurking variables not an issue (if study well done).
Cause-and-effect relationship likely (if experiment well done).
Limitations: Study only looked at people who have had a heart attack, were on heart medication, and looks like all subjects were Italian, and all subjects changed to a healthy diet.
Units of observation: the individuals in the study. Units of analysis: the individuals in the study, as we are comparing the outcomes from each individual, and the outcome from each individual is independent of others.

Answer for Sect. 2.7.4

Quite possibly: confounding! They could have been chosen randomly... though it would be unlikely.

Answer for Sect. 2.7.5

Type of drink: explanatory variable; nominal variable; qualitative variable. Response variable; quantitative variable.

Answers for Sect. 2.7.6

The first design (A).
The third design (C).
Depends on the research focus, but the second probably strikes a balance. The key identifying the source of variation that is likely to be the greatest, and allocate relatively more units there because that is the source of variation that is more important to quantify.
Easiest would be the design using the smallest number of forests (Design A), as collecting data within a forest is likely to be more convenient than moving around to many forests.