Post by anonprof on Apr 5, 2013 13:54:33 GMT -5
I'm a methods person (two degrees in stats, years of teaching grad stats behind me, etc.) and figured it was worth chiming in. I spend hours attempting to make my graphical and tabular displays as efficient and intuitive as possible, and as a reviewer, I expect every author to do the same. It is unacceptable to submit sub-par presentations even at the review stage and it is 100% the author's responsibility to do it right.
I often end up criticizing authors for:
I would much rather have a three page tables or several figures than any one item from the list above. That doesn't mean you should be verbose in your presentation. Minimalism is the way to go, but do present everything important. If it takes you three pages to do it, then acknowledge the reviewer's point and explain why you can't cut any variables in your response letter.
If you really need to cut, then socy users suggestion to present your results in wide format is a good way to go. So, you would have:
I often end up criticizing authors for:
- Omitting standard errors: all this indicates is that the author doesn't understand how statistical inference works or how regression works at very fundamental levels. Please don't do this.
- Omitting control variables from the table: this denies the reader the ability to holistically evaluate the model. Just saying they are included isn't good enough. Their omission also makes replication much more difficult.
- Too many digits: You are almost never justified in using more than 2 decimal places in a table. By the CLT, a sample size of at least 4,000,000 is needed to replicate results to three decimal places, while at least 40,000,000 cases are needed to replicate to four decimal places.
- Lack of alignment: The estimates and standard errors should be aligned at the decimal point. Learn how to use the decimal delimiter in Word or r@{.}l in LaTeX. it makes a huge different in readability.
- Crappy Excel graphs: Put a little time into these or learn R. Here I'm referring to things like angled text on the axes, way too many tickmarks, way too many labels, presenting odds ratios on the linear scale, using the wrong sort of graph, and just making ugly plots. There are dozens of books on effective graphical presentation (Tufte, Cleveland, Wilkinson). Read them. Even Stata or SAS can make halfway decent graphs with enough coercion.
I would much rather have a three page tables or several figures than any one item from the list above. That doesn't mean you should be verbose in your presentation. Minimalism is the way to go, but do present everything important. If it takes you three pages to do it, then acknowledge the reviewer's point and explain why you can't cut any variables in your response letter.
If you really need to cut, then socy users suggestion to present your results in wide format is a good way to go. So, you would have:
Model 1 Model 2
x_1 1.23* (0.12) 1.11* (0.11)
x_2 4.56* (0.13) 4.44* (0.22)
x_3 7.77* (0.33)