minor typos in chapter 4

[rpruim]

• we rely on observational data to help us answer causal questions by implementing a target trial. --> is the target trial implemented or merely designed/specified? On first reading it is unclear to me what the definition of target trial is. Is it the randomized trial we would have done if we could, or is it the modification of the hypothetical randomized trial that we actually implement. It would be nice if this were clearer at first reading.
• Let’s look at an example. I am going to simulate 100 observations. -> elsewhere you use "we" for the author(s).
• Is as section 4.4.2 planned? If not, perhaps 4.4.1 could be subsumed into 4.4 and the title of 4.4 modified.
• In causal inference, we are often interested in marginal effects, mathematically, this means that we want to marginalize the effect of interest across the distribution of factors in a particular population that we are trying to estimate a causal effect for. --> Defining marginal effects by saying that they marginalize seems circular. Perhaps average would work better here? Or a bit more discussion of marginal vs conditional distributions.
• holding all other variables in the model constant. --> This is not a correct interpretation of regression, so best to either avoid it or make it clear that it is incorrect. See for example https://journals.sagepub.com/doi/pdf/10.1177/1536867X1601600103 (Regressions are commonly misinterpreted by David Hoaglin, 2016)
• we often say something like “a one-unit change in the exposure results in a coefficient change in the outcome holding all other variables in the model constant. --> missing closing quotation mark.
• an conditional -> a conditional

[malcolmbarrett]

holding all other variables in the model constant. --> This is not a correct interpretation of regression, so best to either avoid it or make it clear that it is incorrect. See for example journals.sagepub.com/doi/pdf/10.1177/1536867X1601600103 (Regressions are commonly misinterpreted by David Hoaglin, 2016)

I'm finalizing this section now, and I completely disagree with this claim (after reviewing the paper you shared). The model can and does extrapolate over space where we don't have data. That's exactly why we're using a statistical model and not stratification. It's just that some values of "holding constant" are not interpretable in real life. That's why some people, e.g., center on means or modes or whatever. But we're dealing with that in a better way with marginalization throughout the book