Re: SPSS GLM - between * within factor interactions
I'm not sure if this will work for your experiment but have you tried respecifying the model in terms of a nested design. Using syntax you can type A(B) in the DESIGN subcommand which means A nested within B. The code example below would fit the model Y = Constant + A+ B(A) UNIANOVA Y BY A B /METHOD = SSTYPE(3) /INTERCEPT = INCLUDE /CRITERIA = ALPHA(.05) /DESIGN = A B(A) . So you get tests for the main effect of A and the effect of B within A. "Johannes Hartig" <[EMAIL PROTECTED]> wrote in message [EMAIL PROTECTED]">news:[EMAIL PROTECTED]... > Hi group, > I've posted this problem a while ago without getting an answer, and I > still didn't find out myself, > so here is another try: > (I'm working with SPSS 9.0) > When I do a GLM-repeated measures analysis, in SPSSI get tests for all > interactions bewteen every within-and every between-subject factor. > Supposed I don't want to include all these interactions in my model, > e.g. the interaction between a covariate and a within-subject > factor: Is there a way to adjust my model not to include this effect? > (Syntax welcome!) > Or, if this is not possible (I vaguely remember MANOVA in former > versions did not test these interactions as default), is there a good > statistical reason to test all possible within*between-interactions? > Any help is appreciated, > Johannes Hartig > === 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: normality and regression analysis
Mike: It's really the error terms in the regression model that are required to have normal distributions with constant variance. We check this by looking at the properties of the residuals from the regression. You shouldn't expect the response (dependent) variable to have a normal distribution with a fixed mean since then you wouldn't be doing regression. By the way, you have a fine Statistics Department at VPI. I am sure they do excellent consulting. Jon Cryer At 06:39 PM 5/11/00 -0400, you wrote: >I would like to obtain a prediction equation using linear regression for >some data that I have collected. I have read in some stats books that >linear regression has 4 assumptions, 2 of them being that 1) data is >normally distributed and 2) constant variance. In SAS, I have run >univariate analysis testing for normality on both my dependent and >independent variable (n=147). Both variables have distributions that are >skewed. > >For the dependent variable: skewness=0.69 and Kurtosis=0.25. >For the independent variable: skewness=0.52 and Kurtosis= -0.47. > >The normality test (Shapiro-Wilk Statistic) states that both the dependent >and independent variables are not normally distributed. > >I have also transformed the data (both dependent and independent variables) >using log, arcsine, and square root transformations. When I run the >normality tests on the transformed data, the test shows that even the >transformed data is not normally distributed. > >I realize that I can use nonparametric tests for correlation (I will use >Spearman), but is there a nonparametric linear regression? If not, is it >acceptable to use linear regression analysis on data that is not normally >distributed as a way to show there is a linear relationship? > >thanks in advance..Mike > > > > >=== >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/ >=== > > _ - | \ Jon Cryer[EMAIL PROTECTED] ( ) Department of Statistics http://www.stat.uiowa.edu\ \_ University and Actuarial Science office 319-335-0819 \ * \ of Iowa The University of Iowa dept. 319-335-0706\ / Hawkeyes Iowa City, IA 52242FAX319-335-3017 | ) - V === 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/ ===
normality and regression analysis
I would like to obtain a prediction equation using linear regression for some data that I have collected. I have read in some stats books that linear regression has 4 assumptions, 2 of them being that 1) data is normally distributed and 2) constant variance. In SAS, I have run univariate analysis testing for normality on both my dependent and independent variable (n=147). Both variables have distributions that are skewed. For the dependent variable: skewness=0.69 and Kurtosis=0.25. For the independent variable: skewness=0.52 and Kurtosis= -0.47. The normality test (Shapiro-Wilk Statistic) states that both the dependent and independent variables are not normally distributed. I have also transformed the data (both dependent and independent variables) using log, arcsine, and square root transformations. When I run the normality tests on the transformed data, the test shows that even the transformed data is not normally distributed. I realize that I can use nonparametric tests for correlation (I will use Spearman), but is there a nonparametric linear regression? If not, is it acceptable to use linear regression analysis on data that is not normally distributed as a way to show there is a linear relationship? thanks in advance..Mike === 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: test
Jeff Li wrote: > > test I gotta say...I like the rotating ORACLE in the v card. Cheap trick, but very effective made better by keeping it out of the body of the message. Required statistical content: which test? === 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: misusing stats: examples
- Original Message - From: Gene Gallagher > Here is an error that is subtle, but very common. The statistical > test (multiple regression) was applied perfectly, but the > statistical inference was wrong. > My first reference to this type of error is in the classic, > but highly controversial, ecology treatise by Andrewartha & Birch > (1954): The distribution and abundance of animals, p. 580. > > These Australian ecologists wanted to show that animal > populations aren't controlled by density-dependent factors like > competition or predation. They regress 14 years of thrip (an > insect) abundance vs weather variables. They considered weather a > density-independent factor (mortality from a storm or a hot day > isn't directly related to animal density). > They conclude, "...altogether, 78 per cent of the variance > was explained by four quantities which > were calculated entirely from meteorological records. This left > virtually no chance of finding any other systematic cause for > variation, because 22 per cent is a rather small residium to > be left as due to random sampling errors. > All the variation in maximal numbers from year to year may therefore > be attributed to causes that are not related to density: not only > did we not find a "density-dependent factor," but we also showed that > there was no room for one." Also, as both weather and population data tend to be autocorrelated, simple regression would tend to overestimate the significance of correlation (though not the correlation itself), by imputing more degrees of freedom than genuinely present. -Robert Dawson === 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
"Donald F. Burrill" schrieb: > On Wed, 10 May 2000, Johannes Hartig wrote: > > > I guess I have to accept there is no way to customize within * between > > interactions in GLM. Thanks for the tip using regression, but I think > > in future I'd rather try to give a meaning to the default interactions > > ;-) This leads me back to the second part of my original question: Is > > there some good statistical reason *against* removing e.g. a > > covariate * within-factor interaction from a repeated measures model? > > Well, that depends. If the data contain an interestingly strong and > significant interaction, one wouldn't want to remove it, would one? > The existence of such an interaction implies that the slope of the > regression line of (response varible) vs. (covariate) differs from cell > to cell of the design; I'd surely want to examine those differences > before deciding to throw them out. (They might, for instance, be trying > to tell me that I should be dealing with, say, the logarithm or the > square root of the covariate, rather than with the covariate in its > original form.) > On the other hand, if you are determined that the within-cell > regression slope for this covariate will be the same in all cells (that > is, the regression lines will be strictly and exactly parallel throughout > the design), regardless of what the data may be trying to convey, then > removing the interaction will do that. (Doing that will also provide > useful information for comparing the model with interaction to the model > without interaction, of course.) > It is always a fair approach to ask, and then to show, how much > the inclusion (or not) of one or more terms in a model affects the > results of the analysis. Thank you very much for your answer. This is quite what I thought and why I looked for a way to include or remove specific effects. And this is why I'm astonished by the fact SPSS does not allow me to customize within * between factors in its standard procedures. I guess I'll rather try different Software than using regression with dummy variables or something, but thanks to all for the tips! Best wishes, Johannes === 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/ ===
test
This is a multi-part message in MIME format. --4156CD35FFFE2FC4070E5293 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit test --4156CD35FFFE2FC4070E5293 Content-Type: text/x-vcard; charset=us-ascii; name="jeffli.vcf" Content-Transfer-Encoding: 7bit Content-Description: Card for Jeff Li Content-Disposition: attachment; filename="jeffli.vcf" begin:vcard n:Li;Jeff x-mozilla-html:FALSE org:Oracle GZhttp://members.tripod.com/marivera/oracle.gif"> adr:;; version:2.1 email;internet:[EMAIL PROTECTED] title:Asst. Pre-sale Consultant fn:Jeff Li end:vcard --4156CD35FFFE2FC4070E5293-- === 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: misusing stats: examples
Gene Gallagher wrote: > I have recently seen examples of the thrip fallacy in the op-ed > pages of the Boston Globe. Massachusetts has implemented > state-wide standardized testing and has increased state funding > for school districts with low test scores. Statistical analysis > reveals that Five or six socioeconomic factors > (parents educational level, annual salary, % two-parent households, > etc) account for over 90% of the variance in town-to-town K-12 > standardized test scores. The implication is that only 10% of > the variance in mean test scores COULD be due to differences in > curriculum, teacher quality, or financing for the school (Take > that Teacher's Unions!). Some might conclude that spending > money on schools & teachers since only 10% of the town-to-town > variance in these scores could be due to factors outside the home. > This fallacy fails to consider that a high median income and > other socioeconomic factors often are strongly associated with > a better tax base, lower class sizes, better trained teachers, > more innovative curriculum etc. > This fallacy should have a name, but I don't know it. I point > my students to Wright's path analysis and structural > modeling approaches (LISREL, and AMOS) to show alternatives > to the misleading inference based on an R^2 in a multiple > regression equation. There is an additional fallacy here (I think). As I understand it they used town-town means to infer a small effect of other factors on children's education. This is an example of the ecological fallacy. The town mean scores allow no firm inference about the effect of any factor on individual children (they could be similar in magnitude, different in magnitude or even in different directions). Thom === 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: Correlation over time
"G. Anthony Reina" wrote: > I'm looking for a way to show how two continuous signals are correlated > over time. In other words, if x(t) and y(t) are correlated signals (with > some phase lag between them), does that correlation change over time? > (and if so, then how does it vary) Depending on the kind of process you are dealing with the correlation can stay constant or will change during measurements. > > > What I'd ideally like to get is something like the spectrogram except > instead of frequency vs time, the axes would be correlation vs. lag vs. > time. > > The most obvious solution I've thought of is to use a sliding window on > each signal to evalute the cross-correlation (at different lags) over > small epochs. I'm wondering if there are other more elegant solutions > out there. > Most of the programs dealing with signal processing (Matlab, Labview,...) have standard a cross - correlation (Rxy) function. It's not necessary to program it by yourself. If you deal with a non-stationairy process, splitting up the sequence is indeed not a bad idea, put in most of the cases you will be satisfied with the analyse of the total sequence. > > I'd appreciate any advice on the subject. > > Thanks. > -Tony Reina I hope I've helped you a little bit, Koen Maertens === 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/ ===
