Dear R-list members, I compared my statistics with my supervisor yesterday. He is using STATISTICA, I am using R. We both loaded the same data-file and did a two-way ANOVA with treatment and trial as factors; treatment means were then compared to the untreated control using Dunnett's test. Surprisingly, we got different values, especially the F values differed. In the following Dunnett's test, some treatments were significant in R, but not in STATISTICA. We checked the basic settings of the programs and found out, that STATISTICA is using "sigma-restricted parametrization" and "type VI sum of squares" for the calculations, but so far, I could not find any information about the sum of squares type, which is applied in R. Is it possible, that these settings are the reason for the different values we got? And how can I change in R these settings? Here the code I used: bt_assay_devtime_Model <- aov(developmenttime~treatment+run+treatment*run, data=bt_assay) summary(bt_assay_devtime_Model) postHocs_bt_assay_devtime <- glht(bt_assay_devtime_Model, linfct=mcp(treatment="Dunnett"), base=1) summary(postHocs_bt_assay_devtime) confint(postHocs_bt_assay_devtime) Thanks a lot for your help! Simone Eisenring Michael, Msc. PhD Student Federal Department of Economic Affairs, Education and Research EAER Institute of Sustainability Sciences ISS Biosafety Reckenholzstrasse 191, CH-8046 Z?rich Tel. +41 44 37 77181 Fax +41 44 37 77201 michael.eisenring at agroscope.admin.ch<mailto:michael.eisenring at agroscope.admin.ch> www.agroscope.ch<http://www.agroscope.ch/>
Change sum of squares type for ANOVA
4 messages · michael.eisenring at agroscope.admin.ch, Michael Friendly, John Fox +1 more
Use car::Anova() for type II and type III sums of squares and F-tests. The sequential type I tests computed by anova() are rarely sensible
On 5/20/2016 3:11 AM, michael.eisenring at agroscope.admin.ch wrote:
Dear R-list members, I compared my statistics with my supervisor yesterday. He is using STATISTICA, I am using R. We both loaded the same data-file and did a two-way ANOVA with treatment and trial as factors; treatment means were then compared to the untreated control using Dunnett's test. Surprisingly, we got different values, especially the F values differed. In the following Dunnett's test, some treatments were significant in R, but not in STATISTICA. We checked the basic settings of the programs and found out, that STATISTICA is using "sigma-restricted parametrization" and "type VI sum of squares" for the calculations, but so far, I could not find any information about the sum of squares type, which is applied in R. Is it possible, that these settings are the reason for the different values we got? And how can I change in R these settings? Here the code I used: bt_assay_devtime_Model <- aov(developmenttime~treatment+run+treatment*run, data=bt_assay) summary(bt_assay_devtime_Model) postHocs_bt_assay_devtime <- glht(bt_assay_devtime_Model, linfct=mcp(treatment="Dunnett"), base=1) summary(postHocs_bt_assay_devtime) confint(postHocs_bt_assay_devtime) Thanks a lot for your help! Simone Eisenring Michael, Msc. PhD Student Federal Department of Economic Affairs, Education and Research EAER Institute of Sustainability Sciences ISS Biosafety Reckenholzstrasse 191, CH-8046 Z?rich Tel. +41 44 37 77181 Fax +41 44 37 77201 michael.eisenring at agroscope.admin.ch<mailto:michael.eisenring at agroscope.admin.ch> www.agroscope.ch<http://www.agroscope.ch/> [[alternative HTML version deleted]]
Dear Michael and Michael,
I took a quick look at the Statistica website, and their definition of "types" of sums of squares doesn't appear to be the same as the common definition, which originated with SAS. As I read their materials, what they call "Type VI" sums of squares would correspond to the "Type III" tests provided by Anova() in the car package by setting type=3.
As well, the reference to "sigma-restricted coding" in the Statistica documentation corresponds to contrasts provided by contr.sum() in R. This is important for Type III tests because it produces a row-basis for the model matrix with orthogonal effects. One way to do this is via options(contrasts=c("contr.sum", "contr.poly")). See ?Anova for more information.
I hope this helps,
John
-----------------------------
John Fox, Professor
McMaster University
Hamilton, Ontario
Canada L8S 4M4
Web: socserv.mcmaster.ca/jfox
-----Original Message----- From: R-help [mailto:r-help-bounces at r-project.org] On Behalf Of Michael Friendly Sent: May 20, 2016 8:50 AM To: michael.eisenring at agroscope.admin.ch; r-help at r-project.org Subject: Re: [R] Change sum of squares type for ANOVA Use car::Anova() for type II and type III sums of squares and F-tests. The sequential type I tests computed by anova() are rarely sensible On 5/20/2016 3:11 AM, michael.eisenring at agroscope.admin.ch wrote:
Dear R-list members, I compared my statistics with my supervisor yesterday. He is using
STATISTICA, I am using R. We both loaded the same data-file and did a two-way ANOVA with treatment and trial as factors; treatment means were then compared to the untreated control using Dunnett's test. Surprisingly, we got different values, especially the F values differed. In the following Dunnett's test, some treatments were significant in R, but not in STATISTICA. We checked the basic settings of the programs and found out, that STATISTICA is using "sigma- restricted parametrization" and "type VI sum of squares" for the calculations, but so far, I could not find any information about the sum of squares type, which is applied in R. Is it possible, that these settings are the reason for the different values we got? And how can I change in R these settings?
Here the code I used: bt_assay_devtime_Model <- aov(developmenttime~treatment+run+treatment*run, data=bt_assay) summary(bt_assay_devtime_Model) postHocs_bt_assay_devtime <- glht(bt_assay_devtime_Model, linfct=mcp(treatment="Dunnett"), base=1) summary(postHocs_bt_assay_devtime) confint(postHocs_bt_assay_devtime) Thanks a lot for your help! Simone Eisenring Michael, Msc. PhD Student Federal Department of Economic Affairs, Education and Research EAER Institute of Sustainability Sciences ISS Biosafety Reckenholzstrasse 191, CH-8046 Z rich Tel. +41 44 37 77181 Fax +41 44 37 77201 michael.eisenring at agroscope.admin.ch<mailto:michael.eisenring at agroscop e.admin.ch> www.agroscope.ch<http://www.agroscope.ch/> [[alternative HTML version deleted]]
______________________________________________ R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting- guide.html and provide commented, minimal, self-contained, reproducible code.
Dear Michael F What you say is of course true but Michael E (or possibly Simone) seems to be using a function glht from an unnamed package. Since we do not have the data or the offending output it is hard to be sure though. Michael D
On 20/05/2016 13:49, Michael Friendly wrote:
Use car::Anova() for type II and type III sums of squares and F-tests. The sequential type I tests computed by anova() are rarely sensible On 5/20/2016 3:11 AM, michael.eisenring at agroscope.admin.ch wrote:
Dear R-list members, I compared my statistics with my supervisor yesterday. He is using STATISTICA, I am using R. We both loaded the same data-file and did a two-way ANOVA with treatment and trial as factors; treatment means were then compared to the untreated control using Dunnett's test. Surprisingly, we got different values, especially the F values differed. In the following Dunnett's test, some treatments were significant in R, but not in STATISTICA. We checked the basic settings of the programs and found out, that STATISTICA is using "sigma-restricted parametrization" and "type VI sum of squares" for the calculations, but so far, I could not find any information about the sum of squares type, which is applied in R. Is it possible, that these settings are the reason for the different values we got? And how can I change in R these settings? Here the code I used: bt_assay_devtime_Model <- aov(developmenttime~treatment+run+treatment*run, data=bt_assay) summary(bt_assay_devtime_Model) postHocs_bt_assay_devtime <- glht(bt_assay_devtime_Model, linfct=mcp(treatment="Dunnett"), base=1) summary(postHocs_bt_assay_devtime) confint(postHocs_bt_assay_devtime) Thanks a lot for your help! Simone Eisenring Michael, Msc. PhD Student Federal Department of Economic Affairs, Education and Research EAER Institute of Sustainability Sciences ISS Biosafety Reckenholzstrasse 191, CH-8046 Z?rich Tel. +41 44 37 77181 Fax +41 44 37 77201 michael.eisenring at agroscope.admin.ch<mailto:michael.eisenring at agroscope.admin.ch> www.agroscope.ch<http://www.agroscope.ch/> [[alternative HTML version deleted]]
______________________________________________ R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.