affect / effect:
effect (verb): correct if you
can substitute "bring about" for it. "We effected
an improvement in math scores."
affect (verb): "affECT":
correct if you can substitute "alter" for it. "The announcement affected the mood of the group". effect (noun): the result of
doing something. "The announcement had a bad effect on the mood of the group."
affect (noun): "AFFect": psychology
jargon only!, meaning something like the emotional appearance of
someone: "The depressed man exhibited a very flat affect."
- - - - - - - - - - - Format: Even though I had seen a bit of all of the datasets,
most (not all) projects started out with my feeling like I had been
thrown into the deep end of the pool!
You should never assume your reader knows what you're doing till you've
told her!.
So START with a brief description of the dataset (including size=n) and
the variables and what you're going to look at/for! Then the
analysis.
What worked best in terms of clarity and coherence was
"in-line" graphs and tables, or heavily commented-on output. Also
a narrative paperclipped to multiple short, numbered, Appendices of
output, referenced by number in the narrative (read with narrative in
one hand, appendix in the other.)
What worked worst was focusing on the checklist of analyses as
the organization; got in the way of seeing what was in the data.
Experiment, observational study, sample?
It's not an experiment unless you impose a treatment which is
supposed to make the subject change. So measuring IQ, brain
volume, hours of dreaming, etc. may be invasive but they're still just
observational, because you want to see the subject as it is.
Outliers, oddballs: Try to find out or guess who they are,
using the rest of the dataset! May be a cue to why they
are where they are. Missing data: A couple of projects had what was probably
missing data masquerading as numbers. I caught one in the
proposal stage but not the other one. Negative numbers when they
make no sense is one tipoff; a bunch of 99's is another. 0's can
creep in, especially when data is changed from one format to another.
Transformations: Several datasets had relationships that
looked like log or other transformations might give nice relationships;
but actually writing down the relationship as a formula didn't happen
much! Missed opportunities...