Re: differences between groups/treatments ?
Rich Ulrich wrote: These are not quite equivalent options since the first one really stinks -- If you are considering drawing conclusions about causation, you need *random assignment* and the two Groups of performance are the furthest thing from random. Let's see: the simple notion of regression-to-the-mean says that the Best performers should fall back, the Worst performers should improve; that's a weird main-effect, which should wreak havoc with interpreting other effects. Or: If the Pre is powerful enough to measure potential, then a continued-growth model says that Best performers should improve more, even given no treatments. This pattern was described in an obit about two-three years ago in the NY Times. A statistician's obit noted that he'd found a flaw in the Israeli air force's training program. Apparently, the Israeli air force was punishing the worst performers in a test because this usually produced a better performance in subsequent tests and was supposedly much more effective than positive reinforcement. They'd found that positive reinforcement of the best performers often resulted in a poorer performance on the next test. This now-deceased statistician pointed out the confounding effect of regression to the mean on this assessement of negative and positive reinforcement. The effectiveness of negative reinforcement (punishment) could be nothing more than a chance effect. I wish I had the citation for the study or the obit. Does anyone else in the group have a citation of this study? -- Eugene D. Gallagher ECOS, UMASS/Boston Sent via Deja.com http://www.deja.com/ Before you buy. === This list is open to everyone. Occasionally, less thoughtful people send inappropriate messages. Please DO NOT COMPLAIN TO THE POSTMASTER about these messages because the postmaster has no way of controlling them, and excessive complaints will result in termination of the list. For information about this list, including information about the problem of inappropriate messages and information about how to unsubscribe, please see the web page at http://jse.stat.ncsu.edu/ ===
Re: differences between groups/treatments ?
At 04:31 PM 6/22/00 +, Gene Gallagher wrote: This pattern was described in an obit about two-three years ago in the NY Times. A statistician's obit noted that he'd found a flaw in the Israeli air force's training program. Apparently, the Israeli air force was punishing the worst performers in a test because this usually produced a better performance in subsequent tests and was supposedly much more effective than positive reinforcement. They'd found that positive reinforcement of the best performers often resulted in a poorer performance on the next test. This now-deceased statistician pointed out the confounding effect of regression to the mean on this assessement of negative and positive reinforcement. The effectiveness of negative reinforcement (punishment) could be nothing more than a chance effect. A few years ago the journal "Statistical Methods in Medical Research" published an issue on regression to the mean (vol 6, no 2, 1997). It included the five following papers: Regression towards the mean, historically considered (pp. 103-114) M Stigler S. The impact and implication of regression to the mean on the design and analysis of medical investigations (115-128) Chuang-Stein C.,M Tong D. Adjusting for regression toward the mean when variables are normally distributed (129-146) Lin H., Hughes M. Non-normal variation and regression to the mean (147-166) Chesher A. Using regression models for prediction: shrinkage and regression to the mean (167-183) Copas J. Rich Strauss Dr Richard E Strauss Biological Sciences Texas Tech University Lubbock TX 79409-3131 Email: [EMAIL PROTECTED] Phone: 806-742-2719 Fax: 806-742-2963 === This list is open to everyone. Occasionally, less thoughtful people send inappropriate messages. Please DO NOT COMPLAIN TO THE POSTMASTER about these messages because the postmaster has no way of controlling them, and excessive complaints will result in termination of the list. For information about this list, including information about the problem of inappropriate messages and information about how to unsubscribe, please see the web page at http://jse.stat.ncsu.edu/ ===
Re: differences between groups/treatments ?
regression to the mean is not necessarily appropriate when looking at pretest scores ... and then gain or improvement ... if we had parallel tests ... one for pre and one for post ... when nothing happens inbetween ... then maybe so ... please see a short summary of this scenario ... applied to grading based on gain or improvement ... at http://roberts.ed.psu.edu/users/droberts/5501.htm don burrill and i wrote a short paper on this ... those high on the pre CAN gain MORE than those low on the pre IF ... the correlation between pre and post is decent AND, most importantly ... if the variance on the POST is LARGER than the variance on the pre the exact reference to the paper is ... Roberts and Burrill, Spring 1995, Gain score grading revisited, Educational Measurement: Issues and Practices, V14, #1 At 04:31 PM 6/22/00 +, you wrote: Rich Ulrich wrote: These are not quite equivalent options since the first one really stinks -- If you are considering drawing conclusions about causation, you need *random assignment* and the two Groups of performance are the furthest thing from random. Let's see: the simple notion of regression-to-the-mean says that the Best performers should fall back, the Worst performers should improve; that's a weird main-effect, which should wreak havoc with interpreting other effects. Or: If the Pre is powerful enough to measure potential, then a continued-growth model says that Best performers should improve more, even given no treatments. This pattern was described in an obit about two-three years ago in the NY Times. A statistician's obit noted that he'd found a flaw in the Israeli air force's training program. Apparently, the Israeli air force was punishing the worst performers in a test because this usually produced a better performance in subsequent tests and was supposedly much more effective than positive reinforcement. They'd found that positive reinforcement of the best performers often resulted in a poorer performance on the next test. This now-deceased statistician pointed out the confounding effect of regression to the mean on this assessement of negative and positive reinforcement. The effectiveness of negative reinforcement (punishment) could be nothing more than a chance effect. I wish I had the citation for the study or the obit. Does anyone else in the group have a citation of this study? -- Eugene D. Gallagher ECOS, UMASS/Boston Sent via Deja.com http://www.deja.com/ Before you buy. === This list is open to everyone. Occasionally, less thoughtful people send inappropriate messages. Please DO NOT COMPLAIN TO THE POSTMASTER about these messages because the postmaster has no way of controlling them, and excessive complaints will result in termination of the list. For information about this list, including information about the problem of inappropriate messages and information about how to unsubscribe, please see the web page at http://jse.stat.ncsu.edu/ === == dennis roberts, penn state university educational psychology, 8148632401 http://roberts.ed.psu.edu/users/droberts/droberts.htm === This list is open to everyone. Occasionally, less thoughtful people send inappropriate messages. Please DO NOT COMPLAIN TO THE POSTMASTER about these messages because the postmaster has no way of controlling them, and excessive complaints will result in termination of the list. For information about this list, including information about the problem of inappropriate messages and information about how to unsubscribe, please see the web page at http://jse.stat.ncsu.edu/ ===
Re: differences between groups/treatments ?
Look up the topic regression to the mean. This means that of values measured several times , when extremes are revisited they can be at a more typical value. In article 8itf0t$a68$[EMAIL PROTECTED], Gene Gallagher [EMAIL PROTECTED] wrote: Rich Ulrich wrote: These are not quite equivalent options since the first one really stinks -- If you are considering drawing conclusions about causation, you need *random assignment* and the two Groups of performance are the furthest thing from random. Let's see: the simple notion of regression-to-the-mean says that the Best performers should fall back, the Worst performers should improve; that's a weird main-effect, which should wreak havoc with interpreting other effects. Or: If the Pre is powerful enough to measure potential, then a continued-growth model says that Best performers should improve more, even given no treatments. This pattern was described in an obit about two-three years ago in the NY Times. A statistician's obit noted that he'd found a flaw in the Israeli air force's training program. Apparently, the Israeli air force was punishing the worst performers in a test because this usually produced a better performance in subsequent tests and was supposedly much more effective than positive reinforcement. They'd found that positive reinforcement of the best performers often resulted in a poorer performance on the next test. This now-deceased statistician pointed out the confounding effect of regression to the mean on this assessement of negative and positive reinforcement. The effectiveness of negative reinforcement (punishment) could be nothing more than a chance effect. I wish I had the citation for the study or the obit. Does anyone else in the group have a citation of this study? -- Eugene D. Gallagher ECOS, UMASS/Boston Sent via Deja.com http://www.deja.com/ Before you buy. Sent via Deja.com http://www.deja.com/ Before you buy. === This list is open to everyone. Occasionally, less thoughtful people send inappropriate messages. Please DO NOT COMPLAIN TO THE POSTMASTER about these messages because the postmaster has no way of controlling them, and excessive complaints will result in termination of the list. For information about this list, including information about the problem of inappropriate messages and information about how to unsubscribe, please see the web page at http://jse.stat.ncsu.edu/ ===
Re: differences between groups/treatments ?
On 19 Jun 2000 18:01:28 -0700, [EMAIL PROTECTED] (Dónal Murtagh) wrote: ... Firstly, thank you for your comments. Am I right in saying that the two (equivalent) options I have are: These are not quite equivalent options since the first one really stinks -- If you are considering drawing conclusions about causation, you need *random assignment* and the two Groups of performance are the furthest thing from random. Let's see: the simple notion of regression-to-the-mean says that the Best performers should fall back, the Worst performers should improve; that's a weird main-effect, which should wreak havoc with interpreting other effects. Or: If the Pre is powerful enough to measure potential, then a continued-growth model says that Best performers should improve more, even given no treatments. For simple change-scores (and ANOVA interactions) from dichotomous groups, you assume that neither of those possibilities are true, if you want to be able to interpret them. The Regression model at least places the contrasts into the realm of comparing the regression lines. Your fundamental knowledge of what is happening will probably come from examining and comparing the scatterplots, pre-post, for the two treatments. (Another thing to note from the picture: Are there ceiling/basement effects on the performance test?) 1.ANOVA Yijk = mew + Ai + Bj + ABij + Eijk Ai: a fixed factor representing the treatments (2 levels) Bj: a fixed factor representing prior perfromance (2 levels) ABij: an interaction between Ai and Bj Yijk: the score of the kth child who received treatment i and is from group j Eijk: random error I suspect that this model is inapporpriate, as the Eijk term will represent between subjects (children) variation, which is not usually included in the estimate of random error. 2.MLR Y = Bo + B1*X1 + B2*X2 + B3*X3 + E X1: prior performance (0 = weak, 1 = strong) X2: treatment (0 = treament A, 1 = treatment B) X3: treatment*prior performance I appreciate that prior performance is probably better considered as a continuum, rather than a dichotomy. - Treating it as a continuum is better by a lot, even if you were sure that the Performance scale was close to the ANOVA-analytic ideal, a normal distribution. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html === This list is open to everyone. Occasionally, less thoughtful people send inappropriate messages. Please DO NOT COMPLAIN TO THE POSTMASTER about these messages because the postmaster has no way of controlling them, and excessive complaints will result in termination of the list. For information about this list, including information about the problem of inappropriate messages and information about how to unsubscribe, please see the web page at http://jse.stat.ncsu.edu/ ===
Re: differences between groups/treatments ?
On Tue, 20 Jun 2000, Murtagh wrote: Firstly, thank you for your comments. Am I right in saying that the two (equivalent) options I have are: 1.ANOVA Yijk = mew + Ai + Bj + ABij + Eijk Ai: a fixed factor representing the treatments (2 levels) Bj: a fixed factor representing prior perfromance (2 levels) ABij: an interaction between Ai and Bj Yijk: the score of the kth child who received treatment i and is from group j Eijk: random error I suspect that this model is inapporpriate, as the Eijk term will represent between subjects (children) variation, which is not usually included in the estimate of random error. I do not understand this comment. What source(s) of random error exist in this design APART from variation between subjects within cells? Between-subjects variation (as residuals from the model) defines the standard error-variance term against which the variability in the systematic effects is tested. 2.MLR Y = Bo + B1*X1 + B2*X2 + B3*X3 + E X1: prior performance (0 = weak, 1 = strong) X2: treatment (0 = treament A, 1 = treatment B) X3: treatment*prior performance -- hence with the coding shown for X1 and X2, 1 = strong prior performance and treatment B, 0 = all other conditions. And E = Eijk of the ANOVA model. B1 is a straightforward function (depending on the coding of X1, of course) of the Ai in the ANOVA model, B2 of the Bj (and depends on the coding of X2), and B3 of Ai, Bj, and ABij. I appreciate that prior performance is probably better considered as a continuum, rather than a dichotomy. _I_ would consider it so. In fact, the first thing I'd do is ask for scatterplots of post-performance vs. pre-performance for all the cells in the design I was considering. (In what you've described, that's two cells.) THEN decide whether it appeared to make better sense to divide the continuum into two (or more) pieces, or to model it AS a continuum, possibly with non-linear functions. 1. If there are children of different sexes, you may be able to consider a three-way design, although I suspect it would be unbalanced, which (I also suspect!) may induce serious difficulties for you. You mean that there would not be the same numbers in each group? Yes. I can't see why this should cause problems, but then that's probably due to my relative ignorance of linear models! Doesn't cause problems in one-way designs. But in 2-way designs (let alone 3-way, 4-way, ...) unequal n's induce association of some kind between the design factors. People who do multiple regression don't have much problem with this, it's their normal situation; but people who try to do formal ANOVA design-of-experiments (and are therefore accustomed to the notion that the factors are mutually independent (and therefore are orthogonal)) are sometimes boggled by (1) the fact that the sums of squares for the several sources of variation do not simply add to the total sum of squares about the grand mean, or (2) the fact that the sums of squares reported depend on the order in which the factors are considered. And many of the standard packages for doing multi-factor ANOVA use algorithms that require the design to be balanced. (A GLM -- general linear model -- program does not usually have such constraints, and may even produce output patterned after the form of a standard balanced ANOVA, but one needs to be aware of (1) and (2) above.) 2. Your Performance information you have chosen to dichotomize, although it is presumably (quasi-)continuous to start with. You might find out something useful by treating it as a continuous predictor rather than as a dichotomy: in effect carrying out an analysis of covariance with pre-treatment reading score as the covariate, whether you used an "Analysis of Covariance" program or a "Multiple Regression" program or a "General Linear Model" (GLM) program to do the arithmetic. Presumably, this could achieved by simply using the pre-treatment score itself (rather than 0 or 1) for the value of X1 in the suggested MLR model above? Right. And if the pre-post relationship should turn out to be detectably nonlinear, you can substitute some candidate nonlinear function(s) of X1 and see if that helps. There may be nonlinearity to be EXPECTED: in the nature of a reading test, there is a highest possible score (all items right, e.g.) and a lowest possible score (no items right, e.g.). Students who perform well pre-treatment cannot have change scores that would put them above the highest possible score at post-treatment; so it would not be surprising if (a) change correlates negatively with pre-treatment, (b) post scores were censored at the maximum (and negatively skewed), (c) pre scores were censored at the minimum (and positively skewed), and/or (d) the post vs. pre scatterplot showed
Re: differences between groups/treatments ?
On Tue, 20 Jun 2000, Rich Ulrich wrote: On 19 Jun 2000 18:01:28 -0700, [EMAIL PROTECTED] (Dónal Murtagh) wrote: ... Firstly, thank you for your comments. Am I right in saying that the two (equivalent) options I have are: These are not quite equivalent options since the first one really stinks -- Sorry, Rich, I must take issue with yoku. If the first option really stinks, so does the second: they are, in fact, equivalent, as Donal describes the second (with dichotomies for X1 and X2). If you are considering drawing conclusions about causation, This is a fair enough warning, I suppose; but I don't recall reading anything in the original post that implied that it was desired to show causation. (Can't think of anything that expressly denied it either; but I still think you're reading it into, rather than out of, the problem.) you need *random assignment* and the two Groups of performance are the furthest thing from random. For that matter, had it been specified that the treatments were assigned at random? In any case, I'd be interested in knowing how you would propose that performance might be assigned at random. ;-) Let's see: the simple notion of regression-to-the-mean says that the Best performers should fall back, the Worst performers should improve; that's a weird main-effect, which should wreak havoc with interpreting other effects. Or: If the Pre is powerful enough to measure potential, then a continued-growth model says that Best performers should improve more, even given no treatments. Ummm... I think you have to postulate that the POST is powerful enough, unless you're assuming that the Pre and Post measures are identical (which they may be, of course; though that introduces other measurement issues). For simple change-scores (and ANOVA interactions) from dichotomous groups, you assume that neither of those possibilities are true, if you want to be able to interpret them. Only if you want to be able to interpret them SIMPLY. The Regression model at least places the contrasts into the realm of comparing the regression lines. Yes, provided one is modelling the pretest as a continuum rather than as a coded dichotomy, as Donal described it. Your fundamental knowledge of what is happening will probably come from examining and comparing the scatterplots, pre-post, for the two treatments. (Another thing to note from the picture: Are there ceiling/basement effects on the performance test?) Good advice. I concur. - Treating it as a continuum is better by a lot, even if you were sure that the Performance scale was close to the ANOVA-analytic ideal, a normal distribution. Did you mean the ERRORS (or residuals) in the Performance scale, perhaps? -- Don. Donald F. Burrill [EMAIL PROTECTED] 348 Hyde Hall, Plymouth State College, [EMAIL PROTECTED] MSC #29, Plymouth, NH 03264 603-535-2597 184 Nashua Road, Bedford, NH 03110 603-471-7128 === This list is open to everyone. Occasionally, less thoughtful people send inappropriate messages. Please DO NOT COMPLAIN TO THE POSTMASTER about these messages because the postmaster has no way of controlling them, and excessive complaints will result in termination of the list. For information about this list, including information about the problem of inappropriate messages and information about how to unsubscribe, please see the web page at http://jse.stat.ncsu.edu/ ===
Re: differences between groups/treatments ?
On Mon, 19 Jun 2000, Donal wrote: I'm currently analysing data resulting from a study of children's reading ability. I shall resist the temptation to quibble over your inability to observe reading ability (as distinct from some indeterminate lower bound on that ability) ... As you describe the study, you have an unspecified number of children divided into four groups in a two-way design of Treatments (2 levels) by Prior Performance (2 levels). This would naturally lend itself to a two-way analysis of variance, or equivalently (pace Joe Ward) to a multiple regression analysis with three predictors: Treatment, Performance, and Treatment*Performance. If there are indeed effects attributable to Treatment and Performance, this analysis will be more sensitive to them than the two separate t-tests you propose. And if there is an interaction between Treatment and Performance, as there may well be, the sensitivity to possible effects increases. Whether this is the best analysis available is another question entirely. 1. If there are children of different sexes, you may be able to consider a three-way design, although I suspect it would be unbalanced, which (I also suspect!) may induce serious difficulties for you. 2. Your Performance information you have chosen to dichotomize, although it is presumably (quasi-)continuous to start with. You might find out something useful by treating it as a continuous predictor rather than as a dichotomy: in effect carrying out an analysis of covariance with pre-treatment reading score as the covariate, whether you used an "Analysis of Covariance" program or a "Multiple Regression" program or a "General Linear Model" (GLM) program to do the arithmetic. 3. In addition to sex, there may be other lurking variables in your data that could be used as predictors. Whether it is sensible to consider including them in a hypothetical model depends partly on how many children you have all together, and partly on the distribution of any such candidate variable among _these_ children. The study involves two treatments and each child's reading ability was measured before and after the application of one of the treatments. Thus, each child received one or the other (but not both) of two possible treatments. The children are divided into two groups: Well, that's not quite true. You chose to categorize them into two groups, but they could equally well have been divided into three, or four, or six (depending on the number of children available and one's degree of interest in fine-tuning the "Weak/Strong" dimension). And if you have both boys and girls, you have two sexes as well, and it would not be surprising if they differed in their responses to the two treatments. And how about the ages of the children? Weak readers: those whose pre-treatment reading score was less than the mean pre-treatment reading score Strong readers: those whose pre-treatment reading score was greater than the mean pre-treatment reading score It is more usual, in situations like this, to divide at the median rather than the mean. (For one thing, you're more likely to end up with groups of at least approximately equal size.) Did you have a reason for using the mean? Where did you put persons whose score was equal to the mean? Anyhow, I would like to test (for each treatment) whether or not the change in reading score (Post-treatment score - Pre-treatment score) is the same for weak readers and strong readers. I have attempted to test this by: 1. Creating a new variable, "Change" Change = Post-treatment score - Pre-treatment score 2. Using a two-sample t-test to determine whether or not the mean value of "Change" measured over the weak readers is significantly different from the mean value of "Change" measured over the strong readers. Similarly, I'd like to test whether or not the change in the reading score is the same for each treatment. I have attempted to test this by: 1. Creating a new variable, "Change"[as above] 2. Using a two-sample t-test to determine whether or not the mean value of "Change" measured over treatment A is significantly different from the mean value of "Change" measured over treatment B However, I am not certain that this is the best way to test my hypothesis, if anyone can suggest a better way, I'd be very grateful for their assistance. Do these in fact represent your hypotheses, or were they just the closest you thought you could get to what you really wanted to find out? E.g., are you REALLY only interested in the change scores, or are you (perhaps ALSO) interested in the level of proficiency attained, as measured (however imperfectly) by your post-test reading scores? -- DFB. Donald F. Burrill [EMAIL