Archived
E-mail Discussion List
|
|
Return to E-mail Discussion page Next in thread>"If your experiment needs statistics, you ought to have done a better >experiment." > >- Lord RutherfordOooooooh! That kinda talk makes me hoppin' mad. Irrelevant of the (historical) context of Rutherford's statement, variations on those words still haunt the halls of research institutions not to mention the world at large. So, in the interest of dispelling falsehoods, I would like to take on that point of view here. First some definitions: 1. Numbers are used to describe some aspect of physical reality: For example, a chair may be 34 inches tall and weighs 15 lbs. 2. Statistics are numbers which summarize lots of numbers. For example, after making fifty measurements, I calculated that the average amount of time it takes me to get to work is 14 minutes. 3. Statistical tests are mathematical tools which allow us to discriminate between different sets of numbers. For example, a new variety of wheat (3 to 6 tons per acre) produces 10% more grain than an old variety (2.7 to 5.4 tons) (P < 0.001). The probability that the two varieties produce the same amount of grain is less than 0.001. What are the implications of these definitions? First, numbers focus on only one aspect of reality at a time (e.g. height, weight, temperature). So one cannot rely on a number for the whole story. In my example above, what you may be most interested in before buying the chair is the width of the seat. Since statistics are numbers which summarize many numbers, one cannot expect a complete picture from just one statistic. But many statistics, like many short summary statements (e.g. an outline or study guide), can provide a reasonable picture of reality. For example, if I told you that I only walked to work, the single statistic of 14 minutes would probably be fair enought. But if I drove to work on some portion of the days I took measurements, you may not have a clear idea of how far awya from work I live. With any kind of summary, only a small part of the story is told. With the right statistics, a fair description of important characteristics may be presented. Alternatively, with the wrong statistics, an unfair description may be presented. Only critical thinking and an intimate knowledge of the subject matter being described can allow one to determine the difference between a fair and unfair summary. There is natural variation in the world. This is a fact that no one will dispute. For example, not everyone is the same height, nor do crops yield the same amount of grain from year to year. But people want to make decisions in life, and they rely on experience to guide them. In my example, the two varieties did not always produce a set amount of grain. The new variety produced 3 tons in one field and 6 tons in another, and, somewhere in between in ten other fields. But other farmers want to know what the chances are that it will produce more in their fields. No one knows for 100% certainty because of natural variation of soil types, weather, etc. But based on the information provided the new variety is a good bet for most farmers in the long run. As for the "lies, damn lies, and statistics", only the ignorant and uncritical will dismiss all statistics as lies. The knowing and critical will ask "what do these numbers REALLY tell us". And for those scientists who do not approve of experiments that need only one or two measurements (molecular biologists have been very weak thinkers in this area), they will only study the obvious, most of which has been fully documented, and be quickly out of a job. Dr. Brian McSpadden Gardener USDA-ARS Root Disease and Biological Control Unit Washington State University Pullman, WA 99164 (509) 335 1116 (509) 335 7674 FAX |
