There is a great deal already known about age-level-appropriate statistical analysis. Keep it simple at this grade level. The purpose is not to produce an analysis that requires advanced understanding of statistics, but for the *child* to understand, apply, and explain an analysis. Restrict the investigation to 1 experimental variable (the age groups). Forget about interactions, and don't even think of trying to compute a formal confidence interval or to carry out formal statistical tests! That is totally inappropriate at this age. There are other measures that capture the essence of the important statistical ideas, which are much more grade-appropriate.
At this age, if the statistical instruction leading up to this point has been appropriate (it often isn't), a child should have some understanding of the concepts of "central tendency", and "variability" as summarys of data. Measures of central tendency that are accessible to a 6th grader are the median and to some extent, but less so, the mean. Variability is probably only understandable at this point as the range, although the student may also have been introduced to the box plot. In this case he/she may be able to quantify the variability a bit more in terms of where the middle 50% of the observations lie, vs. the more extreme observations. I would suggest trying to get the child to first make a graph with the actual observations plotted. Create 5 age-group categories, ordered along the horizontal axis. For each age-group, plot the actual score of each of the 10 tests on the vertical axis. Then summarize the plot by finding the median for each age group (add a horizontal bar to the plot at that point) and lines extending from the median to the upper and lower extreme within each age group. This is a minimal, but perfectly suitable plot from which to start exploring trends. If the child understands how to construct a box plot, take the plot this one extra step. Box plots for this age typically plot the median, the interquartile range, and whiskers to the extreme points. There is no attempt at this age to plot whiskers to some fraction of the interquartile range, with the extremes specifically noted. For information on statistical analysis for children, see the ASA website (and look for the information on K-6 materials): http://www.amstat.org/education/ql-projects.html ************************************************************************ Ellen M. Wijsman COURIER DELIVERY ADDRESS ONLY: Research Professor Ellen M. Wijsman Div. of Medical Genetics and 1914 N 34th St., suite 209 Dept. Biostatistics Seattle, WA 98103 BOX 357720, University of Washington (Note: Use this address Seattle, WA 98195-7720 EXACTLY as given above, and phone: (206) 543-8987 use ONLY for courier delivery!!!) fax: (206) 616-1973 email: [EMAIL PROTECTED] web page: http://faculty.washington.edu/wijsman ************************************************************************* On Mon, 5 Jan 2004, Thom wrote: > m v wrote: > > > > My wife and I both have degrees in math and have taken undergrad > > statistics courses long ago but are having trouble helping our child > > determine the appropriate statistics to use for her middle school > > science project and how to translate the result into english for her > > conclusion. > > > > Her project was to find if short term memory was better for young > > adults than for younger or older age groups. She devised a memory > > test and tested 10 people in each of five different age groups. > > I think I'd want to know a little about the precise age groups. As far > as STM goes you'd expect young children (say under 7 or 8) to be worse > than young adults and you might expect age-related decline to effect > very old adults (say 70+; though your daughter might have a biased > sample of healthy, active older participants). I think it would be hard > to show effects in between because the differences might be quite small. > Any differences will reflect strategy differences (especially for very > young subjects who don't use efficient memory strategies) as well as STM > 'capacity' per se. Also the materials influence the extent of > differences - some materials will probably show bigger effects (e.g., > nouns, easily nameable pictures) because they make more strategies available. > > Given the target audience I think that confidence intervals and a plot > of means might be appropriate (though not the most powerful test). > Plotting 1.4 standard errors will give you approximately an independent > t test at p<.05 between means. > > Thom > . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
