The following comments, originally drafted in response to the commentaries I graded one particular year, tend to apply more generally and are offered in that spirit.

A good commentary is well-organized, concise, clearly written, and reflective of good understanding of epidemiologic language and concepts. Here are a few issues worth noting.

1. Differentiating between selection bias and generalizability – Selection bias refers to internal validity; generalizability is a matter of external validity. Although there are situations where it can be difficult to distinguish the two, one way to think about the distinction is that findings heavily influenced by selection bias are incorrect. There is no valid interpretation available unless the selection bias is taken into account in the analysis or at least in the interpretation. When the only problem is lack of generalizability, the findings themselves may be correct in their application to the population at which the study was directed. Problems arise from attempts to apply those findings to other populations or situations. In Kleinbaum, Kupper, and Morgenstern's terminology, the criterion for selection bias is whether or not the "actual" population (the population to which the study results actually apply) is different from the "target" population (the population which the investigators intended to study). For example, the fact that a study deliberately limited its enrollment to a narrowly-defined segment of the population does not a cause selection bias, since the inclusion/exclusion criteria define the target population. Such criteria may limit the generalizability (external validity) of the study's findings to other population segments.
2. "Association", "prediction", and "effect" – variables are associated (or, there is an association) when the variables covary in some way, i.e., when the distribution of one of the variables is different for different values of the other variable or variables. This concept is like circumstantial evidence, in that it says nothing about whether the covariation is causal in any sense or in either direction. Variables predict when their values are associated with another variable or variables AND the predictor variables exist (or can be known) before the values of the predicted variable(s). A variable can predict without causing (e.g., a weather forecast). Variables affect (or, variables have an effect) when one variable influences another, not simply predicts its distribution. So "affect" and "effect" should be reserved for use when causal meaning is intended. Note, however, that there are well-established usages in social sciences and applied statistics, in which "predictive" and "effect" refer to statistical relationships (e.g., "main effects") rather than the more specific meanings presented above.
3. "Non-significant" is not equivalent to "not associated" – a comparison for which "p>0.05" or "p=0.20" is one for which the numerical evidence is insufficient to "exclude" random variability as a potential explanation for the observed results. An association may or may not be present in the study data. But the study data do not provide an adequate basis for statistical inference that an association exists in a population from which the data represent a random sample. As a shorthand, the variables in such a comparison are often referred to as being "not associated". But it is important to note that a "non-significant" study may provide little if any evidence against the existence of an association. In such cases one cannot infer, statistically, that an association does not exist, without bringing in additional information (e.g., other studies or a priori probabilities).

In reading through written commentaries over the years, I have noticed a number of grammar, punctuation, and usage problems that occur with some frequency. I offer the following comments by way of general feedback. I must admit, though, that my reference for these comments (Edwin C. Woolley, Franklin W. Scott, and Frederick Bracher, College handbook of composition, 6 ed, Boston, DC Heath, 1958) may no longer be authoritative!

however, . . . – "however" is a "conjunctive adverb" and may not be used to join two independent clauses (clauses capable of constituting a sentence by themselves). So a construction like "The studies were well-conducted, however they suffered from imprecision of measurement." is NOT correct. To use these words in this order requires the use of a semi-colon, a coordinate conjunction (e.g., "and" or "but"), or a period followed by a new sentence.

Restrictive and nonrestrictive modifiers – when should a comma be used – Dependent clauses, participial phrases, and appositives can be used to modify a word in either a nonrestrictive or a restrictive manner. A nonrestrictive modifier adds information but does not limit the scope of the word being modified. Such a modifier can be removed without changing the central meaning of the sentence. A restrictive modifier limits the meaning of the word being modified, so that without the modifier the sentence has a different meaning. Since the same words can serve as a nonrestrictive modifier or a restrictive modifier, the absence or presence of a comma is crucial to signaling which meaning is intended.

To illustrate, here are two sentences, identical except for punctuation:

The meaning of the first sentence, the nonrestrictive construction, would not be seriously changed by removal of the modifying clause "which were conducted in industry". It must be clear from the context which studies are being referred to, and the information about their having been conducted in industry is purely supplementary. The restrictive onstruction, in contrast, limits the scope of "The studies" to the ones that were conducted in industry, leaving open the possibility that studies in other locations did not influence policy.

Placement of "only" – the word "only" should be placed as close as possible to the word or phrase to which it applies, e.g., "The study had only 40 participants.", NOT "The study only had 40 participants." ("only" modifies "40").

et al. is an abbreviation – "et al." is short for "et alli", and so requires a period. It is also borrowed from Latin, so in English writing may be italicized.

"less" versus "fewer" – although this distinction appears to be falling by the wayside, even in material that comes from the University, "fewer" is the correct choice when referring to countable objects, "less" when referring to an amount. For example, a study with fewer subjects has less precision. The express checkout aisle in the supermarket is intended for customers purchasing fewer than 10 items.

Learning to write well takes effort, as does writing well even when one knows how. But the effort is well spent. Good writing makes a good impression. Submitting a document with typos and grammatical errors is llke spilling your soup at a dinner party. Well-written memoranda, proposals, and manuscripts are easier to read and have more impact.

_________________________________________________________