Chemistry Reference
In-Depth Information
•
Making sure that the statistical techniques will answer the critical questions posed
by our research or meet our experimental objectives—too many methods are adapted
from previous research without careful regard for the difference in objectives
Designing experiments more complex than they need to be such that we are not able
•
to generalize the results or determine the next logical objective of our research
Using a technique that none of our reviewers will understand (several years ago,
•
reviewers were unlikely to reject what they did not understand, but now the bias
appears to reject anything they do not understand—another reason for having a
statistician on the advisory committee and as a coauthor on any manuscripts with
novel applications of techniques unfamiliar to our reviewers)
Interpreting a highly significant correlation as implying cause and effect, but
•
such correlations might be the result of the change in one variable causing the
change in a second variable, or a third variable (perhaps unmeasured) causing the
effect in both of the observed variables, or a mere coincidence
Some frequent mistakes made in using mean separation include
•
Analyzing means across one variable and then across the other variable without
partitioning the
SS
(
sum of squares
) which is analogous to spending $50.00 on a
meal at a good restaurant and then taking the same $50.00 to buy groceries (most
of us can only spend the $50.00 once)
Separating the means of each treatment within an interaction to compare all
•
means (yes, we can trick our SAS program into performing that function, but that
does NOT make it valid!)
Mislabeling the legend to state that “All means in the same column with a differ-
•
ent letter are significantly different from each other (
p
< 0.05)” when the proper
terminology should be “All means in the same column with the same letter are
not significantly different from each other (
p
< 0.05)”
See Table
8.3
for illustrations of these types of mistakes.
In many fields, modeling is an excellent tool to determine trends and patterns.
Models can be built to predict effects, but these models must be validated. Statistical
analysis is not the only way to analyze data. Other methods include yes/no answers.
Enzyme kinetics and Arrhenius plots are two other ways of generating data without
using statistical analysis. Most of what we do in food science, however, has enough
biological variability to require statistical analysis.
Interpretation of data should be done in terms of the original objective. Do they
support our hypothesis? If so, how do they support it? If not, why not? What are the
limitations of our data? What are our data trying to tell us? What is the best way to
present these data?
Make sure to consult a good statistics textbook such as Mason et al. (
2003
) , Ott
and Longnecker (
2008
) or the topic used at school. Refreshing ourselves on some
key points in the topic before consulting with a statistician will probably lead to a
more productive session.
RULE # 8
Don't sell your important textbooks. They come in handy later!
