The issue, more than to HLM or not to HLM, is one of the unit of analysis. Assignments were made by school. Effectively, this means that you have a total of eight cases, 4 treatment and 4 control. A multilevel-model would actually increase your power, allowing you to use the d.f. for your individual units where it is appropriate, like in individual level control variables if you plan to include them.
Alas, you still only have 4 treatment and 4 control variables to estimate your program effect. You would not be estimating your higher-level parameters very accurately, and might decide there is no effect when there is. But, as Don said, you have what you have. --- Donald Burrill <[EMAIL PROTECTED]> wrote: > [Sent to B. Olson and to the EdStat list; I am not > a member of the > other lists to whom the original message was > posted.] > > Brittawni, you write (inter alia) "But because I > have only 4 paired > comparisons b/n schools I have no statistical power > and cannot run HLM." > > Can't say as I understand this. How is this > situation different from > having "only" (!) four levels of a design factor in > ANOVA? > > Or perhaps I should ask, What is it that you need > statistical power FOR, > that is not adequately handled in your present > design? > > In general, as regards statistical power, what you > have is what you > have, unless you have the leisure and the resources > to acquire more > data, or to repeat the evaluation on a grander > scale; neither of which > is possible for you, if I am not mistaken. I can > imagine that you might > not be pleased with the results, if the observed > effects are subtle > enough not to be "significant". I'm having > difficulty imagining > characteristics of a computer program designed to > run HLM (by which I > take it you mean "hierarchical linear modeling", and > not a typo for GLM, > or "general linear model"), or of your data, or of > the interface between > them, that would *prevent* your running the program; > but this seems to > be what you claim to be the case. Perhaps you could > clarify the point? > > On Mon, 3 Nov 2003, Brittawni L. Olson > <[EMAIL PROTECTED]> wrote (edited): > > > Greetings, I have a dilemma I am hoping list serve > members can assist > > me with. I am doing a program evaluation wherein > the program was > > given in 4 schools to all 6th graders; then a > booster of that program > > was given to all 7th graders the following year > (same kids basically). > > I also have 4 comparison schools selected based on > demographic and > > geographic similarities b/n populations served by > schools. > > Because the selection unit of analysis is schools > and the evaluation > > data is collected from individual students I have > what would be a > > two-level model. But because I have only 4 paired > comparisons b/n > > schools I have no statistical power and cannot run > HLM. What > > alternative statistical analyses could be done on > these data given the > > concern about unit of analysis and power? > > > > Brittawni L. Olson, Ph.D. Candidate > > Project Director > > Drug Free Communities Support Project > > Lincoln Council on Alcoholism & Drugs, Inc. > > 914 L Street > > Lincoln NE 68508 > > P: 402-475-2695 > > F: 402-475-2699 > > E: <mailto:[EMAIL PROTECTED]> [EMAIL PROTECTED] > > > ----------------------------------------------------------------------- > Donald F. Burrill > [EMAIL PROTECTED] > 56 Sebbins Pond Drive, Bedford, NH 03110 > (603) 626-0816 > . > . > ================================================================= > Instructions for joining and leaving this list, > remarks about the > problem of INAPPROPRIATE MESSAGES, and archives are > available at: > . http://jse.stat.ncsu.edu/ > . > ================================================================= . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
