RE: Confidence intervals
If your purpose is to try and teach students about confidence intervals, then it makes little sense to start out by telling them the counterexamples. I don't start telling students about standard deviations by describing a Cauchy distribution. Now if we are going to do away with confidence intervals because of a few situations (probably contrived) where they don't work, then we need to rewrite a lot of statistics texts. Maybe the qualitative people have the right idea. Don't use numbers at all. Paul R. Swank, Ph.D. Professor Developmental Pediatrics UT Houston Health Science Center -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of Bill Jefferys Sent: Thursday, September 27, 2001 4:00 PM To: [EMAIL PROTECTED] Subject: Re: Confidence intervals In article <000101c14787$f06dcf90$e10e6a81@PEDUCT225>, <[EMAIL PROTECTED]> wrote: #No more than hypothesis tests necessarily tell you when the null #hypothesis #is false. Nothing is certain in statistics but uncertainty. In what way does a CI tell you where the parameter was (your word), if you can see just by looking at the data that it is impossible for the data to lie in the CI? Bill #Paul R. Swank, Ph.D. #Professor #Developmental Pediatrics #UT Houston Health Science Center # #-Original Message- #From: [EMAIL PROTECTED] #[mailto:[EMAIL PROTECTED]]On Behalf Of Bill Jefferys #Sent: Thursday, September 27, 2001 11:31 AM #To: [EMAIL PROTECTED] #Subject: Re: Confidence intervals # # #In article <008201c14763$9392f260$e10e6a81@PEDUCT225>, #<[EMAIL PROTECTED]> wrote: # ##I use to find that students respoded well to the idea that the hypothesis ##test told you, within the limits of likelihood set, where the parameter ##wasn't while confidence intervals told you where the parameter was. # #But confidence intervals do not necessarily tell you where the parameter #was. Jaynes gives an example of a 90% confidence interval, such that you #can see from the data that it is certain that the parameter does NOT lie #in the interval in question. Tom Loredo gives essentially the same #example in # # http://bayes.wustl.edu/gregory/articles.pdf # http://bayes.wustl.edu/gregory/articles.ps.gz # -- Bill Jefferys/Department of Astronomy/University of Texas/Austin, TX 78712 Email: replace 'warthog' with 'clyde' | Homepage: quasar.as.utexas.edu I report spammers to [EMAIL PROTECTED] Finger for PGP Key: F7 11 FB 82 C6 21 D8 95 2E BD F7 6E 99 89 E1 82 Unlawful to use this email address for unsolicited ads: USC Title 47 Sec 227 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ = = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
RE: Confidence intervals
No more than hypothesis tests necessarily tell you when the null hypothesis is false. Nothing is certain in statistics but uncertainty. Paul R. Swank, Ph.D. Professor Developmental Pediatrics UT Houston Health Science Center -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of Bill Jefferys Sent: Thursday, September 27, 2001 11:31 AM To: [EMAIL PROTECTED] Subject: Re: Confidence intervals In article <008201c14763$9392f260$e10e6a81@PEDUCT225>, <[EMAIL PROTECTED]> wrote: #I use to find that students respoded well to the idea that the hypothesis #test told you, within the limits of likelihood set, where the parameter #wasn't while confidence intervals told you where the parameter was. But confidence intervals do not necessarily tell you where the parameter was. Jaynes gives an example of a 90% confidence interval, such that you can see from the data that it is certain that the parameter does NOT lie in the interval in question. Tom Loredo gives essentially the same example in http://bayes.wustl.edu/gregory/articles.pdf http://bayes.wustl.edu/gregory/articles.ps.gz Bill -- Bill Jefferys/Department of Astronomy/University of Texas/Austin, TX 78712 Email: replace 'warthog' with 'clyde' | Homepage: quasar.as.utexas.edu I report spammers to [EMAIL PROTECTED] Finger for PGP Key: F7 11 FB 82 C6 21 D8 95 2E BD F7 6E 99 89 E1 82 Unlawful to use this email address for unsolicited ads: USC Title 47 Sec 227 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ = = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Confidence intervals
I use to find that students respoded well to the idea that the hypothesis test told you, within the limits of likelihood set, where the parameter wasn't while confidence intervals told you where the parameter was. Paul R. Swank, Ph.D. Professor Developmental Pediatrics UT Houston Health Science Center = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
RE: Analysis of covariance
Some years ago I did a simulation on the pretest-posttest control group design lokking at three methods of analysis, ANCOVA, repeated measures ANOVA, and treatment by block factorial ANOVA (blocking on the pretest using a median split). I found that that with typical sample sizes, the repeated measures ANOVA was a bit more powerful than the ANCOVA procedure when the correlation between pretest and posttest was fairly high (say .90). As noted below, this is because the ANCOVA and ANOVA methods are approaching the same solution but ANCOVA loses a degree of freedom estimating the regression parameter when the ANOVA doesn't. Of course this effect diminshes as the sample size gets larger because the loss of one df is diminished. On the other hand, the treatment by block design tends to have a bit more power when the correlation between pretest and posttest is low (< .30). I tried to publish the results at the time but aimed a bit too high and received such a scathing review (what kind of idiot would do this kind of study?) that I shoved it a drawer and it has never seen the light of day since. I did the syudy because it seemed at the time that everyone was using this design but were unsure of the analysis and I thought a demonstration would be helpful. SO, to make a long story even longer, the ANCOVA seems to be most powerful in those circumstances one is likely to run into but does have somewhat rigid assumptions about homogeneity of regression slopes. Of course the repeated measures ANOVA indirectly makes the same assumption but at such high correlations, this is really a homogenity of variance issue as well. The second thought is for you reviewers out there trying to soothe your own egos by dumping on someone else's. Remember, the researcher you squelch today might be turned off to research and fail to solve a meaty problem tomorrow. Paul R. Swank, Ph.D. Professor Developmental Pediatrics UT Houston Health Science Center -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of jim clark Sent: Thursday, September 27, 2001 7:00 AM To: [EMAIL PROTECTED] Subject: Re: Analysis of covariance Hi On 26 Sep 2001, Burke Johnson wrote: > R Pretest Treatment Posttest > R PretestControl Posttest > In the social sciences (e.g., see Pedhazur's popular > regression text), the most popular analysis seems to be to > run a GLM (this version is often called an ANCOVA), where Y > is the posttest measure, X1 is the pretest measure, and X2 is > the treatment variable. Assuming that X1 and X2 do not > interact, ones' estimate of the treatment effect is given by > B2 (i.e., the partial regression coefficient for the > treatment variable which controls for adjusts for pretest > differences). > Another traditionally popular analysis for the design given > above is to compute a new, gain score variable (posttest > minus pretest) for all cases and then run a GLM (ANOVA) to > see if the difference between the gains (which is the > estimate of the treatment effect) is statistically > significant. > The third, and somewhat less popular (?) way to analyze the > above design is to do a mixed ANOVA model (which is also a > GLM but it is harder to write out), where Y is the posttest, > X1 is "time" which is a repeated measures variable (e.g., > time is 1 for pretest and 2 for posttest for all cases), and > X2 is the between group, treatment variable. In this case one > looks for treatment impact by testing the statistical > significance of the two-way interaction between the time and > the treatment variables. Usually, you ask if the difference > between the means at time two is greater than the difference > at time one (i.e., you hope that the treatment lines will not > be parallel) > Results will vary depending on which of these three > approaches you use, because each approach estimates the > counterfactual in a slightly different way. I believe it was > Reichardt and Mark (in Handbook of Applied Social Research > Methods) that suggested analyzing your data using more than > one of these three statistical methods. Methods 2 and 3 are equivalent to one another. The F for the difference between change scores will equal the F for the interaction. I believe that one way to think of the difference between methods 1 and 2/3 is that in 2/3 you "regress" t2 on t1 assuming slope=1 and intercept=0 (i.e., the "predicted" score is the t1 score), whereas in method 1 you estimate the slope and intercept from the data. Presumably it would be possible to simulate the differences between the two analyses as a function of the magnitude of the difference between means and the relationship between t1 and t2. I don't know if anyone has done that. Best wishes Jim =
RE: if A and B are independent implies A and (A&B) are independent?
if A&B occurs, hasn't A occurred? Paul R. Swank, Ph.D. Professor Developmental Pediatrics UT Houston Health Science Center -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of Nathaniel Sent: Monday, September 17, 2001 12:38 PM To: [EMAIL PROTECTED] Subject: if A and B are independent implies A and (A&B) are independent? Hello, I've got a question. If A and B are idependent does it imply that event A and (A&B) are independent too. My reasoning is: If A and (A&B) are idependent then: P(A|(A&B))=P(A), but P(A|(A&B))=P(A&(A&B))/P(A&B)=P(A&A&B)/P(A&B)=P(A&B)/P(A&B)=1. What does it mean? Could anybody explain me this. Thank you very much Nathaniel = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ = = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
RE: effect size/significance
Dennis said other than being able to say that the experimental group ... ON AVERAGE ... had a mean that was about 1.11 times (control group sd units) larger than the control group mean, which is purely DESCRIPTIVE ... what can you say that is important? However, can you say even that unless it is ratio scale? OTOH, there is a two standard deviation difference, which is large enough to practically hit you over the head. The significance test is more important when the effect is smaller because small effects are much more likely to be chance events. Paul R. Swank, Ph.D. Professor Developmental Pediatrics UT Houston Health Science Center = = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
RE: Recoding Binary variables in STATA
What software are you using? Paul R. Swank, Ph.D. Professor Developmental Pediatrics UT Houston Health Science Center -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of Riddler Sent: Tuesday, September 11, 2001 6:31 AM To: [EMAIL PROTECTED] Subject: Recoding Binary variables in STATA Would like to know a quick way of changing many dichotomous variables coded as: 1=Yes 2=No 9=Missing into 1=Yes 0=No .=Missing I found the way to replace '9' with '.' for a list of variables (mvdecode), however, I haven't found how recode the other values to 0 and 1, other than doing a 'replace' statment for each variable. Is there anyway of recoding multiple variables at once? Mark Riddle, MD, MPH&TM = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ = = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
RE: Definitions of Likert scale, Likert item, etc.
In general, as these things always seem to go, many folks call any item from a summated rating scale a Likert item. But, like the use of relationship instead of relation, it seems to be a difficult practice to stamp out. Paul R. Swank, Ph.D. Professor Developmental Pediatrics UT Houston Health Science Center -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of John Uebersax Sent: Thursday, September 06, 2001 12:44 PM To: [EMAIL PROTECTED] Subject: Definitions of Likert scale, Likert item, etc. A recent question made me realize the extent of ambiguity in the use of "Likert scale" and related terms. I'd like to see things be more clear. Here are my thoughts (I don't claim they are correct; they're just a starting point for discussion). Concise responses are encouraged. If there are enough, I'll post a summary. 1. "Likert scaling" strictly refers to the scaling method developed by Likert in the 1930's. If refers entire process of scaling a set of many items (i.e., as an alternative to Thurstone scaling). One step of this is administering many items to individuals. Each item has integer-labeled rating levels. Likert used the method only for attitude measurement, and with response categories indicating levels of agreement to specific statements, like: "I believe the work week should be reduced to 32 hours." 1. strongly disagree 2. mildly disagree 3. neither agree nor disagree 4. mildly agree 5. strongly agree 2. A "Likert scale", strictly speaking, refers to a set of many such items. 3. I do not know if Likert also used a visual analog format such as: neither strongly mildly agree normildly strongly disagree disagree disagree agree agree 1 2 3 4 5 +-+--+--+-+ 4. It seems reasonable to refer to a single such item as a "Likert item." However, many people seem to refer to a single item of this type as a "Likert scale"; that would seem to invite confusion, as Likert's original intent was to produce a scale compused of many such items. 5. Many researchers use such items outside the area of attitude measurement; it seems reasonable to refer to such items as "Likert-type items", to distinguish them from strict Likert items as described above. If anyone has any definitive references that clarify this, I would greatly appreciate learning of them. John Uebersax, PhD (805) 384-7688 Thousand Oaks, California (805) 383-1726 (fax) email: [EMAIL PROTECTED] Agreement Stats: http://ourworld.compuserve.com/homepages/jsuebersax/agree.htm Latent Structure: http://ourworld.compuserve.com/homepages/jsuebersax Existential Psych: http://members.aol.com/spiritualpsych = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ = = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
RE: Bimodal distributions
A bomodal distibution is often thought to be a mixture of two other distibution with different modes. If the distributions have different sizes, then it is possible to have two or more "humps". I once read somewhere (and now can't remember where) that this may be referred to as bimodal (or multimodal). In the bimodal case, some refer to the higher "hump" as the major mode and the other as the minor mode. Paul R. Swank, Ph.D. Professor Developmental Pediatrics UT Houston Health Science Center -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of Wuensch, Karl L. Sent: Thursday, August 30, 2001 11:54 AM To: edstat (E-mail) Subject: Bimodal distributions Does a bimodal distribution necessarily have two modes? This might seem like a silly question, but in my experience many folks apply the term "bimodal" whenever the PDF has two peaks that are not very close to one another, even if the one peak is much lower than the other. For example, David Howell (Statistical Methods for Psychology, 5th, p. 29) presents Bradley's (1963) reaction time data as an example of a bimodal distribution. The frequency distribution shows a peak at about 10 hundredths of a second (freq about 520), no observations between about 18 and 33 hundredths, and then a second (much lower) peak at about 50 hundredths (freq about 25). + Karl L. Wuensch, Department of Psychology, East Carolina University, Greenville NC 27858-4353 Voice: 252-328-4102 Fax: 252-328-6283 [EMAIL PROTECTED] http://core.ecu.edu/psyc/wuenschk/klw.htm = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ = = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
RE: Min n CFA clarification
Yes, there are 42 lambdas since 3 are fixed to one, 45 theta deltas, and 6 phis, three variances, three unique covariances. Paul R. Swank, Ph.D. Professor Developmental Pediatrics UT Houston Health Science Center -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of Marianne and Dimitrios Sent: Saturday, August 18, 2001 9:50 AM To: [EMAIL PROTECTED] Subject: Min n CFA clarification Let me clarify my previous posting. I want to do a confirmatory factor analysis to validate a questionnaire. There are 45 questions (subjects answer using a 1-5 scale). Theoretically, there are 3 subscales with 15 items on each. In a CFA, that gives me 3 factors, 45 error terms, 14 factor loadings on each of 3 factors, and 3 covariances. I figure that gives me 93 parameters. That's the part I need somebody to verify for me. Have I counted the number of parameters correctly? If so, then I should have at least a 1:10 and at best a 1:20 ratio of subjects to parameters. Hence, the estimate of 930-1860 subjects. Is this correct? Thanks to all who are helping me with this. Marianne = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ = = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
RE: Categorical data Take 2
If your going to use discriminant analysis you will need a lot of data and it does assume the predictors are multivariate normal. Generalized linear models would seem best, particularly in the event that you don't know if they are ordinal. You can do a multinomial followed by a cummulative logit model to see if the data are approximately ordinal. Paul R. Swank, Ph.D. Professor Developmental Pediatrics UT Houston Health Science Center -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of Silvert, Henry Sent: Wednesday, August 15, 2001 10:36 AM To: 'Melady Preece'; [EMAIL PROTECTED] Subject: RE: Categorical data Take 2 Why not a discriminant analysis? You might want to develop profiles of people who go into the 5 different success categories -- although they might all be equally sucess except for the last one. Henry M. Silvert Ph.D. Research Statistician The Conference Board 845 3rd. Avenue New York, NY 10022 Tel. No.: (212) 339-0438 Fax No.: (212) 836-3825 > -Original Message- > From: Melady Preece [SMTP:[EMAIL PROTECTED]] > Sent: Wednesday, August 15, 2001 10:57 AM > To: [EMAIL PROTECTED] > Subject: Categorical data Take 2 > > The discussion of categorical data has got me thinking about a project I > am about begin. The goal is to use a variety of individual predictors > (IQ, previous work experience, education, personality) to develop a model > to predict "success" after a vocational rehabilitation program for > psychiatric patients. > > The problem is how to define success. The current data provides 5 > possible outcomes: Full-time employment, part-time employment, ongoing > education, volunteer work, or no change. Clearly there is no argument to > be made that these are linear, but even ordinal is questionable. > > I had thought of using a number of logit regression analyses for the > various outcomes. Or, to use linear regression, rescaling as number of > hours per week employed, and combining the two employment outcomes; > scaling education training in terms of program length; combining volunteer > and no change into number of hours involved in productive activity. That > would give be three outcomes. > > Any suggestions would be much appreciated! > > Melady Preece, Ph.D. > = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ = = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
RE: Interclass Correlation??
If your interest is reliability then you don't need to do any statistical "comparisons". What you are describing is a case for generalizability theory in which you use the data to estimate the variance components and then estimate what the reliability would be if you vary the number of trials. Books by Brennan and Shavelson & Webb or the original by Cronbach et al would be helpful. Cronbach, L., Gleser, G., Nanda, H. & Rajaratnam, N. (1972). The dependability of behavioral measurements. New York: Wiley. Brennan, R. (1983). Elements of generalizability theory. Iowa City, IA: American College Testing Program. Shavelson, R. $ Webb, N. (1991). Generalizability Theory: A primer. Newbury Park, CA: Sage. Paul R. Swank, Ph.D. Professor Developmental Pediatrics UT Houston Health Science Center -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of Clark Dickin Sent: Monday, July 23, 2001 10:08 PM To: [EMAIL PROTECTED] Subject: Interclass Correlation?? I am trying to determine the reliability of a balance test for individuals with Alzheimer's disease. The test involves six different conditions, with each condition consisting of three trials (6 x 3). Each individual has performed the complete test twice, which gives me 6 trials for each of the 6 conditions. I would like to determine at what point the individuals performance becomes reliable (stable). Specifically I want to know how many trials need to be performed in order to determine when the individual has move beyond learning and into actual performance. Specifically, my questions are: (1) whether or not an ICC is the appropriate test to perform, (2) if the ICC is appropriate do I need to calculate an ICC for each set of two consecutive trials or for the entire group of 6 trials for each condition of the six condition test, and (3) Do I need to correct the alpha level to accommodate for the multiple comparisons (.05/# of contrasts)? Any help would be appreciated. Clark Dickin [EMAIL PROTECTED] = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ = = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
postdoc
I am posting this to several lists for maximum coverage. My apologies if you receive more than one copy. The Division of Developmental Pediatrics is accepting applications for a post-doctoral fellowship in quantitative methods with emphasis or interest in psychometric issues. The successful candidate will work with a diverse team of psychologists, physicians, and other health care professionals on a wide variety of behavioral science projects related mainly to development in children. One such project is the creation of a Center to study psychometric issues in the assessment of children particularly for longitudinal research. Methodological areas of focus are psychometric models of growth, particularly Rasch models, longitudinal data analysis, mixed models, and structural equation modeling. Requirements include a Doctoral degree in quantitative methods, psychometrics, or statistics and experience with SAS. The successful candidate will collaborate with investigators on existing projects and may develop an independent research program in a related area. Salary for the position will be $39,500 with excellent fringe benefits. Accepting applications immediately with position to begin no later than September 1. Send vita and three letters of reference to: Dr. Susan Landry Department of Pediatrics University of Texas Houston Health Science Center 7000 Fannin, Suite 2401 Houston, Texas 77030 Or e-mail to: [EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]> Paul R. Swank, Ph.D. Professor Developmental Pediatrics UT Houston Health Science Center = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: convergent validity
You could do this best with a confirmatory factor analysis which would allow you to "compare" coefficients by restricting them to be equal. The measurement problem would be an issue for some programs although MPlus allows factor analysis with ordered categorical responses. It would require you having a sufficient sample size, but that would depend on the number of parameters in the model which is a function of the number of items and the number of expected latent variables. The typical minimum number is 200 although I have seen some papers with fewer than that. At 08:30 PM 3/29/01 +0200, you wrote: >Hi Statisticians, > >First of all, sorry for posting my question in 3 groups, but I'm a bit >of a newby here and I can't find out what the difference is (where can I >read the Charters, or whatever it's called?). > >I would love to have some help on the following: > >I have 2 questionnaires assessing (physical and emotional) health of >heart patients. The 1st measures present state and it's assessed before >treatment and a couple of months after treatment, so that difference >scores can be calculated. The 2nd questionnaire is assessed after >treatment only, and asks respondents how much they have changed on every >aspect (same aspects as the first questionnaire) since just before >treatment. >Respondents received both questionnaires. Now I would like to >investigate the convergent validity of the two domains assessed with >both questionnaire versions. Is there a standard, straightforward way of >doing this? Someone advised me to do a factoranalysis (PCA) (on the >baseline items, the serially measured change scores and the >retrosepctively assessed change scores) and then compare the >factorloadings (I assume after rotation? (Varimax?)). I haven't got a >good feeling about this method for two reasons: >- my questionnaire items are measured on 5- and 7-point Likert scales, >so they're not measured on an interval level and consequently not >(bivariate) normally distributed; >- I have no idea how to compare the factorloadings. Could I calculate >confidence intervals for the loadings? (If yes: how?) > >Thanks in advance for any help or references. > >Heike > > > >= >Instructions for joining and leaving this list and remarks about >the problem of INAPPROPRIATE MESSAGES are available at > http://jse.stat.ncsu.edu/ >= > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: statistical errors
However, if you took the same number of observations from a skewed distribution and tried to use a test of normality to demonstrate a lack of normality, it probably wouldn't be significant. It's always dangerous to try and draw conclusions from very small amounts of data. At 01:56 PM 3/24/01 -0500, Dr. Rich Einsporn wrote: >At 12:16 PM 3/22/01 -0700, Harold W Kerster wrote: >> Maybe the most common mistake is omission of graphic eye-balling. > >Another common error is drawing inferences from graphs! (re: P. Swank's comments below) > >In particular, I think that using graphs to check normality based on small samples is as questionable as formal tests. I have an exercise that I use with my classes to illustrate this: I randomly generate nine data sets of n=10 observations from a normal population and make histograms for each set. I give the nine graphs to the students and tell them that they represent samples from nine different populations. The students are then asked to identify which of the nine populations are normal. Out of 70 students that I have tried this with so far, only two have seen through my ploy and have >correctly picked all nine. The rest have selected no more than 2 of the nine as coming from normal populations. Even my faculty colleagues have been tricked! > >Rich Einsporn >U. of Akron > > >> >On Thu, 22 Mar 2001, Paul Swank wrote: >> > >> >> I couldn't help wanting to add my own 2 cents to the discussion about statistical errors because I have always thought that people put too much faith in formal tests of assumptions. When the tests of assumptions are most sensitive to violations is when they are of less concern, when the sample size is large. When the ramifications of violating assumptions are greatest, when samples are small, the tests have no power to detect violations. There is no substitute for examining your data. If the data are badly skewed, you don't need a normality test to tell you that, a simple histogram will do it. >> >> >> >> > > > >= >Instructions for joining and leaving this list and remarks about >the problem of INAPPROPRIATE MESSAGES are available at > http://jse.stat.ncsu.edu/ >= > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: statistical errors
I prefer the ocular test myself. At 12:16 PM 3/22/01 -0700, Harold W Kerster wrote: > Maybe the most common mistake is omission of graphic eye-balling. > >On Thu, 22 Mar 2001, Paul Swank wrote: > >> I couldn't help wanting to add my own 2 cents to the discussion about statistical errors because I have always thought that people put too much faith in formal tests of assumptions. When the tests of assumptions are most sensitive to violations is when they are of less concern, when the sample size is large. When the ramifications of violating assumptions are greatest, when samples are small, the tests have no power to detect violations. There is no substitute for examining your data. If the data are badly skewed, you don't need a normality test to tell you that, a simple histogram will do it. >> >> >> >> ---- >> >> Paul R. Swank, PhD. >> >> Professor & Advanced Quantitative Methodologist >> >> UT-Houston School of Nursing >> >> Center for Nursing Research >> >> Phone (713)500-2031 >> >> Fax (713) 500-2033 >> >> >> = >> Instructions for joining and leaving this list and remarks about >> the problem of INAPPROPRIATE MESSAGES are available at >> http://jse.stat.ncsu.edu/ >> = >> > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: power,beta, etc.
How about np, for no publication. At 10:55 PM 3/3/01 -0500, you wrote: >when we discuss things like power, beta, type I error, etc. ... we often >show a 2 by 2 table ... similar to > > null truenull false > >retain correct type II, beta > >reject type I, alpha power > > >i think that we need a bit of overhaul to this typical way of doing things ... > >1. each cell needs to have a name ... label ... that reflects the >consequence of the decision (retain, reject) that was made > >i propose something along the lines of > > null true null false > >retaintype I correct, 1C type II error, 2E > >rejecttype I error, 1E type II correct, 2C > > >then, we have names or symbols for probabilities attached to each cell > > null true null false > >retain WHAT NAME/SYMBOL FOR THIS??beta > >reject alpha power > > >DOES ANYONE HAVE SOME SUGGESTION AS TO HOW THE UPPER LEFT CELL MIGHT BE >REFERRED TO via A SYMBOL??? OR, SOME NAME THAT IS DIFFERENT FROM POWER BUT >... STILL GIVES THE FLAVOR THAT A CORRECT DECISION HAS BEEN MADE (better >than making an error)? > >2. i think it would be helpful to first identify each cell with a >distinctive label ... describing the decision (correct, error) and ... the >type ... 1 or 2 > >3. i think it would be helpful to have a system where there are names for >EACH cell (why should the poor upper left be "left" out in the cold??) ... >FIRST ... then some OTHER name/symbol for the probability associated with >that cell > >confusions that might be avoided would be like: > >a. saying type II error is the same as beta ... >b. saying that power is NOT a name for a decision but, rather, THE >probability of making some particular decision > >we have special names for errors of the first and second kind type I >and type II ... and we have symbols of alpha and beta to represent their >associated probabilities > >we have power which is supposed to be the probability of making a certain >kind of decision ... but, no special name for THAT cell like we have given >to differentiate the two kinds of errors one can make ... > >any support out there to try to right this somewhat ambiguous ship? >== >dennis roberts, penn state university >educational psychology, 8148632401 >http://roberts.ed.psu.edu/users/droberts/drober~1.htm > > >= >Instructions for joining and leaving this list and remarks about >the problem of INAPPROPRIATE MESSAGES are available at > http://jse.stat.ncsu.edu/ >= > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: Cronbach's alpha and sample size
The effect of N on alpha is minimal unless the assumptions for alpha are not met. If you have a multidimensional construct then the alpha will tend to go down as the sample size decreases. At leaset I have observed this in monte carlo analyses. At 12:08 PM 2/28/01 +0100, you wrote: >How is Cronbach's alpha affected by the sample size apart from questions >related to generalizability issues? > >Ifind it hard to trace down the mathmatics related to this question >clearly, and wether there migt be a trade off between N of Items and N >of sujects (i.e. compensating for lack of subjects by high number of >items). > >Any help is appreciated, > >Thanks, Nico >-- > > >= >Instructions for joining and leaving this list and remarks about >the problem of INAPPROPRIATE MESSAGES are available at > http://jse.stat.ncsu.edu/ >========= > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: On inappropriate hypothesis testing. Was: MIT Sexism & statistical bunk
I remember a question from some stat book about a situation where there were 8 members of a group, three men and five women (or the reverse, I can't remember which) and on some issue the vote was five to three with all five women voting for. The question was "How likely was this event to occur by chance"? Can we not ask that question? At 05:34 PM 2/15/01 GMT, you wrote: >Rich: > >To be blunt, although >your comments in this forum are often >valuable, you fell far short of two >cents worth this time. > >This is not a popularity contest, it is a statistical >argument. You offered an unsupported >opinion with only one content-related >comment. Let's cut to the chase. > >Please define precisely what you meant in >the phrase > >> - and if you want to know something about how unlikely it was to >>get means that extreme, you can randomize. Do the test. > >a. You do *have* means "that extreme." > >b. There is no "likelihood" to be considered, because >the entire population is available. We were assessing the >original MIT conjecture that to imply there were important >performance differences between male and female biologists >AT MIT would be "the last refuge of the bigot." > >So, my countercomments to you are: > >1. Rather than snipping the Gork example, deal with it. Explain, >in detail, why the Gork women shouldn't be paid more than the men. >My prediction: you can't, and you won't. > >2. You talk about "how unlikely it was." Unlikely when? >Unlikely under what conditions? > >3. (if you choose to answer question 2) Why would the Gork society be >interested in assessing any such likelihood, if they have a >meritocracy, and their only interest lies in assessing whether male >and female Gorks have shown productivity differences? > >If you can actually answer such questions, rather than rendering >an unsupported opinion, you might have two cents worth to add. > >All the best, > >Jim > >- >James H. Steiger, Professor >Dept. of Psychology >University of British Columbia >Vancouver, B.C., Canada V6T 1Z4 > >Comments reflect my opinion only, >- > >On Thu, 15 Feb 2001 10:39:45 -0500, Rich Ulrich <[EMAIL PROTECTED]> >wrote: > >>I am just tossing in my two cents worth ... >> >>On Thu, 15 Feb 2001 07:53:13 GMT, Jim Steiger, posting as >>[EMAIL PROTECTED] (Irving Scheffe) wrote: >> >>< snip, name comment > >> >>> 2. I tried to make the Detroit Pistons example as obvious as I could. >>> The point is, if you want to know whether one population performed >>> better than another, and you have the performance information, [under >>> the simplying assumption, stated in the example and obviously not >>> literally true in basketball, that you have an acceptable >>> unidimensional index of performance], you don't do a statistical test, >>> you simply compare the groups. >> >> >>> >>> Your question about the randomization test seems >>> to reflect a rather common confusion, probably >>> deriving from some overly enthusiastic comments >>> about randomization tests in some >>> elementary book. >> >> - If you are willing, perhaps we could discuss the textbook >>examples. I don't remember seeing what I would call >>"overly enthusiastic comments about randomization." >>When I looked a few years ago, I did see one book with an >>opposite fault, exemplified in a problem about planets. >>I thought the authors' were pedantic or silly, when they refused >>to admit randomization as a first step of assessing whether there >>*might* be something interesting going on. >> >>>Some people seem to >>> emerge with vague notions that two-sample randomization tests make >>> statistical testing appropriate in any situation in which you have >>> two stacks of numbers. That obviously isn't true. >>> Your final question asks if "statistical tests" be appropriate >>> even when not sampling from a population. In some sense, sure. But not >>> in this case. >> >>I can't say that I have absorbed everything that has been argued. >>But as of now, I think Gene has the better of it. To me, it is not >>very appropriate to be highly impressed at the mean-differences, >>when TESTS that are attempted can't show anything. The samples >>are small-ish, but the means must be wrecked a bit by outliers. >> >
Re: ANOVA : Repeated Measures?
Whether or not to use random effects should depend on whether you wish to generalize the results to some populations that the sample is (hopefully) representative of. Usually we wish to generalize to some population of subjects. Typically (but not neccesarily) we are not interested in generalizing to a population of treatments that our current treatments represent. The measure issue is also more commonly seen as fixed rather than random. I am assuming you have some interest in comparing measures to each other (assuming they are comparable) rather than considering this design a multivariate one. The latter case would give rise to a fourth possibility of a two way multivariate anova rather than a three way univariate anova. In either case, I suspect you want subjects random and treatments fixed. At 04:17 PM 2/9/01 GMT, you wrote: > >We have data from an experiment in psychology of hearing. There are 3 >experimental conditions (factor C). We have collected data from 5 >subjects (factor S). For each subject we get 4 measures of performance >(M for Measure factor) in each condition. What is the best way to >analyse these data? > >We've seen these possibilities : > >a) ANOVA with repeated measures with 2 fixed factors : subjects & >conditions and the different measures as the repeated measure factor >(random factor). > >b) ANOVA with two fixed factor (condition & measure) and a random >factor (repeated measure-> subject factor). > >c) ANOVA with one fixed factor (condition) and the other two as >random. > >We think that the a) design is correct (assuming and verifying that >there is no special effect of the measure factor such as training >effects). > >Other psychologist advised us to use the b) design because >psychologists use to consider the subject effect as random. (in >general experiments in psychology are ran with at least 20 to 30 >subjects). > >The last design (c)) is a possibility if we declare that we have no >hypothesis on the effects of subject & repetition factors. > > >I have only little theoretical background in stats and I like to know >what exactly imply these possible designs. > >Thanks in advance for your help > >Sylvain Clement >"Auditory function team" >Bordeaux, France > > > > >= >Instructions for joining and leaving this list and remarks about >the problem of INAPPROPRIATE MESSAGES are available at > http://jse.stat.ncsu.edu/ >= > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: Normality assumption for ANOVA (was: Effect statistics for non-normality)
At 08:29 AM 1/19/01 -0500, Bruce Weaver wrote: >On 17 Jan 2001, Robert J. MacG. Dawson wrote: snip > >Dr. Dawson has touched on something here that I've always found a bit >puzzling--the oft stated ANOVA assumption that the populations from which >you sample must be normal. I've always had a bit of trouble seeing why >that is the case. I'll try to explain why by approaching it gradually. > >Everyone agrees, I think, that if you have a population of scores that is >normally distributed, a z-score calculated as X-Xbar/SD can be referred to >a table of the standard normal distribution. > >If the "population" from which I pull a score is a population of sample >means (i.e., the sampling distribution of the mean), I simply change the >formula for z or t to: > > X-bar - mu(X-bar) >z or t = - > SE(X-bar) > >This is the z or t-test for a single sample. Provided that the sampling >distribution of X-bar is normal (or near enough), I can still refer z to >the standard normal, or t to the appropriate t-distribution. Now this is >where the CLT comes into play. It gives the conditions under which the >sampling distribution of X-bar is normal (or near enough): > > end snip An interesting example I show my students is to take progressively larger samples from a skewed distribution and observe an approximation of the sampling distribution of X bar, which becomes downright beutiful at relatively reasonable sample sizes and then show them the approximate sampling distribution of S, the sample standard deviation, Which requires a much larger sample size to straighten out. As William Ware said, it's the denominator. Your formula above needs to indicate that it is the estimate of the standard error in the denominator. Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: Raschmodelling
It is also possible to simulate rasch modeling using a mixed models approach with a binomial distribution and logit link function. At 10:17 PM 1/17/01 +0100, you wrote: >Maybe they have something for you here: http://www.gamma.rug.nl >They have a couple of programs on Rasch modelling and some other Item >Respons Theory programs. I'm not sure what free demo's they have... > >Sky > >Mag. Stefan Höfer heeft geschreven in bericht ><000e01c07f9e$fe326610$[EMAIL PROTECTED]>... >>I`d like to obtain information about computersoftware (free, trail or >>commercial) to calculate rasch modell. >> >>Stefan >> >> >> >>= >>Instructions for joining and leaving this list and remarks about >>the problem of INAPPROPRIATE MESSAGES are available at >> http://jse.stat.ncsu.edu/ >>= > > > > >= >Instructions for joining and leaving this list and remarks about >the problem of INAPPROPRIATE MESSAGES are available at > http://jse.stat.ncsu.edu/ >= > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 Ý'structions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ Ý=
Re: regression to the mean
divide at the midpoint of the pretest to form two equal size groups. At 01:37 PM 1/17/01 -0500, you wrote: >At 12:28 PM 1/17/01 -0600, Paul R Swank wrote: >>But if you group the subjects on the basis of their pretest scores, the >>lowest group gains 23.1 points while the highest group only gains 19.2. >>Looking at the graph, I note that the person who scored 34 on the pretest >>did not increase as much as I might expect while the person who scored 10 >>increased more than I might expect. The best fit line is obviously flatter >>than the best fit l;ione for the pre and post tests. > > >not sure what you are referring to as the low and high ... but the data >below has been sorted on the pretest from high to low ... and, then i >looked at the mean gain for the low 6 (on the pre) and the top 6 (on the >pre) and found .. > > >> >if i sort the pre from high to low and then list the gain ... we can see >> >easily that the high pres gain more in fact, the top 6 gain about 51 points >> >on average ... while the low 6 gain only about 35 points on average > >thus, unless you did some quite different calculations ... i don't see >where you get a gain of 23 for the low and 19 for the top? > >whenever one discussed RTM ... they seem to talk about the upper level >group (near the top) and the lower level group (near the bottom) and that >is exactly what i did > > > > >= >Instructions for joining and leaving this list and remarks about >the problem of INAPPROPRIATE MESSAGES are available at > http://jse.stat.ncsu.edu/ >= > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: regression to the mean
54 > 18 2457 > 19 2446 > 20 2444 > 21 2441 > 22 2254 > 23 2250 > 24 2255 > 25 2231 > 26 2142 > 27 2032 > 28 2024 > 29 1439 > 30 1143 > >if you are thinking about regression to the mean in the typical way ... how >come this "regression reversal" seems to have occured? > > > > >= >Instructions for joining and leaving this list and remarks about >the problem of INAPPROPRIATE MESSAGES are available at > http://jse.stat.ncsu.edu/ >= > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: Raschmodelling
Try this site. Winsteps is available commercially but bigsteps, the dos precurser, is available free. http://www.winsteps.com/winsteps.htm At 10:30 AM 1/16/01 +0100, Mag. Stefan Höfer wrote: >I`d like to obtain information about computersoftware (free, trail or >commercial) to calculate rasch modell. > >Stefan > > > >= >Instructions for joining and leaving this list and remarks about >the problem of INAPPROPRIATE MESSAGES are available at > http://jse.stat.ncsu.edu/ >= > -------- Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 Ý'structions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ Ý=
mcas
snip The local schools are already being forced to teach to the test. I reviewed my older daughter's science text and thought it was apalling. There would be a 10-page section mediocre discussion of pressure in the ocean and atmosphere, followed by an inane 10-p discussion of pressure in the blood system. There was little to unite the two concepts in that both dealt with a term called pressure that was very poorly described. I told her teacher that I didn't envy him having to teach with a book that was structured so poorly. He said the book was the best of a bad lot and that they chose it over one that they preferred because the content was closer to that being tested with the MCAS. He said that their previous model was that earth sciences were dealt with in a unified package in one year, followed by the life sciences in the years before and after. However, the MCAS tests both earth and life sciences in one exam, so they couldn't go a year without covering both with the same text. I fear that decisions like this are being made state-wide. -- Eugene D. Gallagher ECOS, UMASS/Boston * And get use to it. This is the way it is done in Texas and I expect there to be a big push out of Washington in the near future for everyoone to move in this direction. -------- Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: MA MCAS statistical fallacy
Robert: Why would you expect a strong correlation here? You're talking about tests done a year apart with some new kids in each school and some kids who have moved on. Is regression toward the mean causing all of the noted results. Probably not. But it is quite conceivable that it could be partially responsible for the results. At 03:21 PM 1/11/01 -0400, you wrote: > > >Paul R Swank wrote: >> >> Regression toward the mean occurs when the pretest is used to form the groups, which it appears is the case here. > > Of course it "occurs": - but remember that the magnitude depends on >r^2. In the case where there is strong correlation between the pretest >and the posttest, we do not expect regression to the mean to be >particularly significant. > > Now, it is generally acknowledged that there are some schools which >_consistently_ perform better than others. (If that were not the case, >nobody would be much surprised by any one school failing to meet its >goal!) Year-over-year variation for one school is presumably much less >than between-school variation. > > Therefore, I would not expect regression to the mean to be sufficient >to explain the observed outcome (in which "practically no" top schools >met expectations); and I conclude that the goals may well have been >otherwise unreasonable. Indeed, requiring every school to improve its >average by at least two points every year is not maintainable in the >long run, and only justifiable in the short term if there is reason to >believe that *all* schools are underperforming. > > -Robert Dawson > > >= >Instructions for joining and leaving this list and remarks about >the problem of INAPPROPRIATE MESSAGES are available at > http://jse.stat.ncsu.edu/ >= > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: MA MCAS statistical fallacy
Regression toward the mean occurs when the pretest is used to form the groups, which it appears is the case here. At 08:31 AM 1/11/01 -0400, you wrote: > > >Gene Gallagher wrote: >> >> Those familiar with "regression to the mean" know what's coming next. >> The poor schools, many in urban centers like Boston, met their > ^ >> improvement "targets," while most of the state's top school districts >> failed to meet their improvement targets. > > Wait one... Regression to the mean occurs because of the _random_ >component in the first measurement. Being in an urban center is not part >of the random component - those schools' grades didn't improve because >some of them woke up one day and found that their school had moved to a >wealthier district. > If the effect of nonrandom components such as this is large enough >(as I can well believe) to justify the generalization highlighted above, >and if there was a strong pattern of poor-performing schools meeting >their >targets and better-performing schools not doing so, we are looking at >something else - what, I'll suggest later > > >> The Globe article describes how superintendents of high performing >> school districts were outraged with their failing grades, while the >> superintendent of the Boston school district was all too pleased with >> the evaluation that many of his low-performing schools had improved: >> >> [Brookline High School, for example, with 18 National Merit Scholarship >> finalists and the highest SAT scores in years, missed its test-score >> target - a characterization blasted by Brookline Schools Superintendent >> James F. Walsh, who dismissed the report. > > There *is* a problem here,but it's not (entirely) regression >to the mean. If I recall correctly, Brookline High School is >internationally >known as an excellent school, on the basis of decades of excellent >teaching. >If it couldn't meet its target, it's not because its presence among the >top >schools was a fluke in the first measurement - it's probably because the >targets for the top schools were unrealistic. > > Was there any justification for the assumption voiced by the Boston >superintendant that the top-performing schools were in fact not >performing at >their capacity and would be "smug" if they assumed that their present >per- >formance was acceptable? The targets described seem to imply that no >school >in the entire state - not one - was performing satisfactorily, even the >top >ones. Perhaps this was felt to be true, or perhaps it was politically >more >acceptable to say "you all need to pull your socks up" than to say "the >following schools need to pull their socks up; the rest of you, steady >as she goes." > > As a reductio ad absurdum, if this policy were followed repeatedly, >it would be mathematically impossible for any school to meet its target > every year. That - and not regression to the mean - is the problem >here, I >think. > > -Robert Dawson > > >========= >Instructions for joining and leaving this list and remarks about >the problem of INAPPROPRIATE MESSAGES are available at > http://jse.stat.ncsu.edu/ >= > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: Matrices and Regression Analysis
Your instructor is giving you the problem in this form in the hopes that you will understand what you're doing. The idea is not to look for shortcuts but to work it out. For example, you have all the information in your previous post to calculate the regression coefficients. The variance of the residuals can be determined from the regression equation. At 08:41 PM 12/11/00 GMT, you wrote: >Hello. I was just wondering if there is some sort of shortcut procedure that >you use when multiplying out matrices. My professor has mentioned something >about lamba, lamba prime? I'm pretty much lost on the subject. >Here's an example: >We want to find (x'x)^-1 (x-prime x inverse) >X is given as [1 24 > [1 22 > [1 15 > [1 15 > [1 17 > [1 18 > [1 24 > [1 22 > [1 10 > [1 21 > [1 15 > [1 20 >So, the transpose of x, x' is [1 1 1 1 1 1 1 1 1 1 1 1 > [24 22 15 15 17 18 24 22 10 21 15 20 >Now, I know I have to find the product of x and x'. My question is, is there a >quicker way to do this then the standard way? It is quite time consuming to >do, especially given the time constraints of an exam > > >= >Instructions for joining and leaving this list and remarks about >the problem of INAPPROPRIATE MESSAGES are available at > http://jse.stat.ncsu.edu/ >= > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: how to calculate stand. err. if sample =50% of population
the estimate of the standard error for a sample from a finite population is
[{sigma/sqrt(n)}{sqrt[(N-n)/(N-1)]}].
At 12:39 PM 11/30/00 GMT, [EMAIL PROTECTED] wrote:
>Hello,
>
>who knows an answer to the following question?
>Let's assume that I take a sample of e.g. 100 people. I ask them a
>question and, e.g. 50% say "yes". I construct a 90%-confidence interval
>and get a standard error of 11.6. Fine.
>
>However, this assumes that the population size is unlimited.
>If the population was 100, then I would not have any standard error.
>However, what if the population size is 200? How do I construct a
>confidence interval then?
>
>Anybody with an idea?
>
>Thanks in advance
>
>Nils Riemenschneider
>
>
>
>Sent via Deja.com http://www.deja.com/
>Before you buy.
>
>
>=
>Instructions for joining and leaving this list and remarks about
>the problem of INAPPROPRIATE MESSAGES are available at
> http://jse.stat.ncsu.edu/
>=========
>
Paul R. Swank, PhD.
Professor & Advanced Quantitative Methodologist
UT-Houston School of Nursing
Center for Nursing Research
Phone (713)500-2031
Fax (713) 500-2033
= Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: part-time temporary lectureship at Boston Univ
And to think, some places pay only 1000-1500 per course. At 05:30 PM 8/7/00 -0500, Petr Kuzmic wrote: > > >Mark Glickman wrote: >> >> The Boston Univ Department of Mathematics and Statistics is >> seeking a part-time temporary lecturer to teach an >> introductory statistics class in the Fall. Responsibilities >> include lecturing three times per week (Mon, Wed and Fri >> 2:00-3:00pm), holding office hours, and creating homeworks >> and exams. There will be a grader and graduate teaching >> assistant who will work with the lecturer. The salary >> for the course is $5000. > >$5000 - 25% federal + stat tax = $3750 >16 week semester, that's $234/week after taxes > >How many hours in the week for a University lecturer: > >3 hours / week in the classroom >6 hours / week preparing & reviewing lecture >3 office hours / week >2 hours / week writing homework assignments >1 hour / week grading homeworks w/ graders or reviewing grades >1 hour / week (prorated) writing exams >1 hour / week (prorated) grading exams w/ grader & TA >3 hours / week answering email from students >1 hours / week supervision of TA & grader >1 hour / week general administrative duties >--- >22 hour/week >Indeed it's a half time job (50% appointment, essentially). >So now, $234/week after taxes makes $10.60/hour. > > > > >Interesting. That's about a buck fifty per hour more than a janitor >makes elsewhere in Boston: > >http://careerpath.boston.com/hw.vts?Action=Search&QueryText=VdkVgwKey%3d/ve rity/data/helpwanted/wraps/jobs.918&ResultTemplate=careers/details.hts > >However, the janitorial job required three years of experience, so it's >probably tougher to get. [;)] > > "Vivat Academia, > Vivat Professores!" > > - Petr > > >= >Instructions for joining and leaving this list and remarks about >the problem of INAPPROPRIATE MESSAGES are available at > http://jse.stat.ncsu.edu/ >= > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: sample size program for regression
Do you mean for power analyses? I have a sas program that computes power for various sample sizes and effect sizes for varying numbers of predictors. If that will help, I'll be happy to send it. At 03:19 PM 6/23/00 CDT, you wrote: >Does anybody know of a asmple size program for regressoin? I seem to >remember a program called R2, but a search of the Web didn't turn up >anything. > >Also, does anybody know of a sample size program for logistic regression? > >TIA > >Karen Scheltema, M.A., M.S. >Statistician >HealthEast >1700 University Ave W >St. Paul, MN 55104 >(651) 232-5212 fax: (651) 641-0683 > > >Get Your Private, Free E-mail from MSN Hotmail at http://www.hotmail.com > > > >=== >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/ >======= > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 === 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: Comments about my syllabus
I found your syllabus to be very ambitious for undergraduates. Is this their first stat course? At 07:34 AM 6/18/00 -0400, SM wrote: >Howdy, >I am not a subscriber of this listserv, but was invited to post by E. >Jacquelin Dietz, editor of THE JOURNAL OF STATISTICS EDUCATION. > >I am a social worker (MSW with a Ph.D. in Sociology) and I teach two >sections of statistics (to social work and criminal justice majors) at a >small college in rural North Carolina. I've completed seven statistics >courses on the Ph.D. level. However, my Ph.D. experience with statistic >courses may not have prepared to teach this course to social work >majors. > >I have shared my syllabus with my social work colleagues, but they have >less of a background in teaching statistics than I do! I am interested >in sharing my syllabus with others who teach statistics and get >feedback. > >Two issues that may not be clear on the syllabus: > >1) I prohibit students from using a computer until they have solved the >equation by hand first. I have discovered that students do much better >on exams when they have done the math. For example, I can ask non math >questions on an exam, and students do better. They seem to have a >deeper understanding. Have you experienced this? > >2) Students seem to understand basis statistical concepts when I repeat >the explanation 3 to 5 times in different ways. I use links on my >syllabus, lecture, films (AGAINST ALL ODDS), the text, and supplemental >readings. > >My syllabus can be found at >http://www.uncp.edu/home/marson/360_summer.html . I would appreciate >your guidance, but try not to hurt my feelings! > >Cordially, > >Steve > >Stephen M. Marson, Ph.D., ACSW >Professor/Director, Social Work Program >UNC-P > > > > > >=== >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/ >=== > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 === 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: MANOVA
If some of your independent (predictor) variables are categorical, then the dots and zeros are not a problem but merely a refelction of what would be redundant parameters. At 01:00 AM 6/15/00 +0100, HAideren wrote: >Hi, > >I have run a MANOVA and in the 'Parameter Estimates' section of the results, >some of the cells are filled with a zero or a dot (.). Is there a way to >overcome this problem? If no, should I run a different multivariate test and >what would be the appropriate substitute test? > >Cheers. > > > > > >=== >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/ >======= > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 === 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: attitudes obsolete?
Perhaps you should check out the mesa site at U Chicago, http://mesa.spc.uchicago.edu/ for another take on things. Actually I have studied psychophysical scaling, have read Likert, Thurstone, and Guttman's original papers, and learned from another classic text, Guilford, 1954. Of course, the point is to scale people on attitudes but if yoy wish to have a scale that is not sample dependent and has better psychometric properties. While I have long used summated rating scales I am beginning to appreciate the advantages of some of the new-fangled, more old fashioned methods of scaling. At 02:15 PM 5/17/00 -0400, you wrote: >i guess i don't see it exactly like this ... attitudes have never been >about stimuli ... but people ... people have attitudes ... stimulus objects >don't ... > >in edwards book, which by the way is perhaps the best (so what if it is >old?) book on this topic ... he quotes thurstone (paraphrasing) as >defining an attitude as: '... any symbol, phrase, slogan, person, >institution, ideal or idea WHICH PEOPLE CAN DIFFER ... WITH RESPECT TO >POSITIVE AND NEGATIVE AFFECT ... ' > >even if we scale the items ... ultimately, what we are interested in is >whether people would agree or disagree with some characterization OF a >stimulus ... that is, would see the stimulus in a negative or positive light > >and, even with rasch modeling (and fancier models) ... yes we do scale >items in the sense (cognitive tests for example) in terms of difficulty >BUT, the goal is to be able to use those items to scale the people ... on >some trait of interest > >i wish more people would have a good study of edwards classic book in this area > >allen edwards, techniques of attitude scale construction, 1957, >appleton-century-crofts, inc > >and also actually read likert's work ... to see what he said and did not >say ... rensis likert, a technique for the measurement of attitudes, >archives of psychology, #140, june 1932 ... > >but that's just my take on things > > > >At 11:52 AM 5/17/00 -0500, Paul R Swank wrote: >>The methods of attitude scale construction have gone full circle it seems. >>The original work (Thurstone) evolved out of psychophysical scaling where >>the stimuli were scaled first. Then came Likert with summated rating scales >>that were much easier to construct because the items did not have to be >>scaled first. Now we have Rasch modeling and IRT where we are scaling >>stimuli first again. These procedures are much more mathematically complex >>than the original (psychophyical) ones because now, with the aid of >>computers, we can do it. > >Dennis Roberts, EdPsy, Penn State University >208 Cedar Bldg., University Park PA 16802 >Email: [EMAIL PROTECTED], AC 814-863-2401, FAX 814-863-1002 >WWW: http://roberts.ed.psu.edu/users/droberts/drober~1.htm >FRAMES: http://roberts.ed.psu.edu/users/droberts/drframe.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/ >=== > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 === 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: obsolete methods?
The methods of attitude scale construction have gone full circle it seems. The original work (Thurstone) evolved out of psychophysical scaling where the stimuli were scaled first. Then came Likert with summated rating scales that were much easier to construct because the items did not have to be scaled first. Now we have Rasch modeling and IRT where we are scaling stimuli first again. These procedures are much more mathematically complex than the original (psychophyical) ones because now, with the aid of computers, we can do it. At 11:21 AM 5/17/00 GMT, you wrote: >I have been looking for resources on attitude scale construction. The >methods I have been looking at are things like paired comparisons and >successive intervals. The strange thing about finding descriptions of >these methods is that the only book I can find in print is *Techniques >of Attitude Scale Construction* by Edwards (1957?). In fact, it seems >that nearly all the standard references on these statistical methods >were published in the fifties or before. > >Does anyone know what happened? Did these methods go out of style >bacause they were superceded? > >Regards, >Tom > > >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/ >======= > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 === 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: SPSS GLM - between * within factor interactions
You could also try SAS. At 02:14 PM 5/9/00 -0400, you wrote: >On Tue, 9 May 2000, Johannes Hartig wrote: > >> I have tried modifying the syntax, but I'm not getting any further. >> The within- and between-subject effects are defined seperately >> in /WSDESIGN and /DESIGN, and mixing them only gives me >> cryptic error messages. Could it be possible to customize within * >> between interactions with /LMATRIX or /KMATRIX? I am >> checking already the syntax guide, but no success so far :( >> Thanks for any advice, >> Johannes >> > >How about generating your own dummy variables for the various main >effects and interactions of interest (including dummy variables for >subject), and using REGRESSION instead of GLM repeated measures? You can >use the /TEST subcommand to compare the full model to various reduced >models to produce tests for the main effects and interactions of >interest. For a between-within design, subject will be nested in the >between subjects variables, so I think you'll have to enter those >between subjects variables on one step, and the dummy variables for >subject on the next step. (If you enter the dummy variables for subject >first, you won't be able to enter the between Ss variables, because >they'll provide no further information. It would be like entering codes >for City, and then trying to enter codes for country: Once you know the >city, you already know country.) > >Good luck. >Bruce > > > >=== >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/ >=== > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 === 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: effect size
since t = (M(1) - M(2)) / S*sqrt(1/n(1) + 1/n(2)) and d = (M(1) - M(2)) / S, Doesn't that make d / sqrt(1/n(1) + 1/n(2)) = t ? At 05:51 PM 4/19/00 -0400, you wrote: >is there a standard error ... for an effect size? > >as an example ... say you were looking at differences between means between >control and treatment ... and, the effect size came out to be ... for sake >of argument ... .3 ... in favor of the treatment > >is there (in this case) some standard error ... that could be used to judge >this value in terms of sampling error? and if so, how would one go about >calculating it? > > > >=== >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/ >======= > Paul R. Swank, PhD. Professor & Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 === 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: split half reliability
At 12:55 PM 4/19/00 +1000, you wrote: >Paul R Swank wrote: >> >> High alpha can be obtained when not all items are highly intercorrelated >> with all the other items but it requires having enough items. Lack of item >> homogeneity will certainly be greater problem with short scales. With >> respect to the Spearman-Brown, I don't recommend it. I prefer the >> Guttman-Flanagan split half which is just a special case of alpha for two >> items. >> >I would agree with everything you've said here, but I still would press >you to explain what, exactly, you mean, when you use the term >"homogeneity". Item homogeneity means essentially tau equivalent items. That is, each item has expected value of tau plus a constant. Thus, they are all measuring the same construct. >I'm not familiar with the Guttmann-Flanagan formula, and I'd be pleased >if you would describe it. From your description, it sounds like the >following: >rel = 1 - H/T >where H = sum of the variances of the two split half scores >T = variance of total test scores. >Is that correct? Exactly. The Guttman-Flanagan split half requires the halves to be essentially tau equivalent rather than parallel as the Spearman-Brown correction requires. > >The question I raised earlier about how one splits the scale into halves >still remains. And your response raises another question: why do you >prefer G-F to Sp-Brown? What is the criterion for "better" here? Because of its reduced assumptions (essentially tau equivalent halves instead of parallel) means the GF split half is not inflated. That is, GF < SB. They are only equal when the variances of the halves are equal. Again, having essentially tau equivalent halves is somethimes easier to justify than essentially equivalent items. > >I'll be away from work for the next eight days, back on Friday 28 April. > >Best wishes, thanks for the discussion, >Paul Gardner > >Attachment Converted: "d:\paul\email\attach\Paul.Gardner2.vcf" > Paul R. Swank, PhD. Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 === 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: split half reliability
I have never said that split half is better in general. In general, alpha is probably better. But in those instances when alpha doesn't fit, the split half technique may be quite useful. Coefficent alpha is based upon a model like all of statistics. If the model is inappropriate, then some other model should be used. Of course, we could use alternate forms or test-retest but they have obvious shortcomings as well. The problem with using alpha when it is not appropriate is that, being an average, it doesn't tell you the range. We all know that the mean is only half the story. A random split compared to alpha gives you more information than either one alone. At 03:53 PM 4/18/00 -0400, you wrote: >Trying to have it both ways, >on 18 Apr 2000 08:13:08 -0700, [EMAIL PROTECTED] (Paul R >Swank) wrote: > >> Depends on whether you consider a lack of item homogeneity as unreliability >> or not. If your content is supposed to be homogeneous then a lack of >> homogeneity implies your test has problems. If your content is not >> necessarily homogeneous then the reduced alpha appears to say that your >> test is unreliable when that's not actually the case. All reliability >> coefficients suffer from the same problem. They are all sensitive to a lack >> of reliability and something else. Test retest is sensitive to trait >> instability, alternate forms is sensitive to a lack of parallelness, split >> half is sensitive to the bad split, and alpha is sensitive to lack of item >> homogeneity. It's like any statistical model. If the model isn't >> appropriate, the result is misleading. >> >So, Paul S. is claiming that split-half is better because it is not >precise? Or, are we supposed to carefully select the two halves so >they will match, for our split-half reliability? -- that is the one >strategy that might be unbeatable, if you really have a directions on >how to split the scale. But Paul does not say that much, so far as I >can tell. If you don't do that, it seems that it *ought* to be >impossible to favor one accidental split-half, compared to, >"Using alpha, which gives an average of all split-half results." > >By the way, the standardized item-alpha (provided, for instance, by >SPSS Reliability procedure) is computed STRICTLY from the >correlations, with no reference at all to variances. Usually, that >should be an optimistic estimate. > >Paul Gardner laid out most of the relevant questions, quite well. > >-- >Rich Ulrich, [EMAIL PROTECTED] >http://www.pitt.edu/~wpilib/index.html > Paul R. Swank, PhD. Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 === 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: split half reliability
Depends on whether you consider a lack of item homogeneity as unreliability or not. If your content is supposed to be homogeneous then a lack of homogeneity implies your test has problems. If your content is not necessarily homogeneous then the reduced alpha appears to say that your test is unreliable when that's not actually the case. All reliability coefficients suffer from the same problem. They are all sensitive to a lack of reliability and something else. Test retest is sensitive to trait instability, alternate forms is sensitive to a lack of parallelness, split half is sensitive to the bad split, and alpha is sensitive to lack of item homogeneity. It's like any statistical model. If the model isn't appropriate, the result is misleading. At 07:47 PM 4/17/00 -0400, you wrote: >At 04:26 PM 4/17/00 -0500, Paul R Swank wrote: >>I disagree with the statement that the split-half reliability coefficient >>is of no use anymore. Coefficient alpha, while being an excellent estimator >>of reliability, does have one rather stringent requirement. The items must >>be homogeneous. > >i don't ever seem to recall the coefficient alpha .. REQUIRES homogeneous >content ... but rather, the SIZE of it will be impacted BY item homogeneity > > > > > >=== >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/ >======= > Paul R. Swank, PhD. Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 === 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/ ===
split half reliability
I disagree with the statement that the split-half reliability coefficient is of no use anymore. Coefficient alpha, while being an excellent estimator of reliability, does have one rather stringent requirement. The items must be homogeneous. This is not always the case with many kinds of scales, nor should it be. In many cases homogeneity of item content may lead to reduced validity if the consruct is too narrowly defined. Screening measures often have this problem. They need to be short but they also need to be broad in scope. Internal consistency for such scales would suffer but a split half procedure, which is much less sensitive to item homogeneity, would fit the bill nicely. Paul R. Swank, PhD. Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 === 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: What to do about "simple" techniques
I am new to the list so I am jumping into the middle of this. However, we have to start teaching hypothesis testing somewhere. Even if it goes the way of the Edsel, it will be a slow death because many of us will continue to use when we feel it is appropriate to the question. However, I tell my students that there are always more complicated ways to do things. In many instances these take more math ability, computer skills, or time than I have to explain them. So I am going to show them a method that will work although it won't necessarily be the most powerful or efficient. The key is to first understand inference thoroughly before jumping into more complicated things. You don't teach a first grader multiplication before they understand addition and I don't teach a nursing student a logistic regression before they can understand a chi-square goodness of fit. If we had the ability to teach all the students what we thought they needed to know, we might do things a little differently. Someone mentioned Joe Ward. His colleague Earl Jennings once told me that the biggest impediment to understanding linear models was learning the t test, anova, and regression techniques separately. When I teach linear models I try to get the students to unlearn a lot of what they know. It seems a waste of time but we are not always in charge of the curriculum. When you are given 3 hours to teach a student something about statistics, do you start with linear models? Probably not. Well, I did not intend this to be quite so long so I'll shut up. At 06:05 PM 4/7/00 -0600, you wrote: >Dear all, > >I am interested in what others are doing when faced with techniques that >appear in standard textbooks that are "simpler" (either computationally >and/or conceptually) than better (but more difficult) techniques. My >concern is when the "superior" techniques is either inaccessible to the >audience (for instance, a "stat" 1011 class) or would take considerably >longer to teach (and the semester isn't long enough now) or requires use >of the computer for almost any sample. Some examples of techniques that I >see in lots of stat textbooks but would rarely be used by a >statistician are: 1) chi-square goodness of fit to test for normality >(when Shapiro-Wilk is much better for the univariate case and the >Henze-Zirkler for the multivariate case); 2) paired sample t-tests >(usually better options here such as ANCOVA); 3) sign test (randomization >tests are much superior). I'm sure I left out/didn't think of plenty of >other cases. My question to the group, as someone at the beginning a >career teaching statistics, is what to do? Should some of these tests be >left out (knowing the students may run into the tests in future course work >or in some research? Should the better procedures always be taught, >knowing that the additional difficulty due to level of >mathematics/concepts/computational load may well lose many students? I >don't know yet; What do you thing? > > >___ >Christopher Mecklin >Doctoral Student, Department of Applied Statistics >University of Northern Colorado >Greeley, CO 80631 >(970) 304-1352 or (970) 351-1684 > > > > >=== >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/ >=== > Paul R. Swank, PhD. Advanced Quantitative Methodologist UT-Houston School of Nursing Center for Nursing Research Phone (713)500-2031 Fax (713) 500-2033 === 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/ ===
