help me
Locked
 
|
Dear all,
I have a database with 600 subjects who carried out, a secondary statistical analysis with logistic regression with 5 independent variables. As I can estimate the statistical power to conclude that my results are valid.
Maga |
|
|
Maga,
It sounds like you did a logistic regression with 5
predictors (independent variables) using data from a sample with an N of 600.
You now want to compute power to show that your results are valid. I
want to make two comments. One comment is about the power computation
specifically. First of all, you can not do what you want using spss. You
have to find something else. Do a search on 'statistical power logistic regression'
or a set of search terms like that. One link is to a program called G*Power 3.
I've
never used this program, which seems to be free, but i think it could be a
good option for you. The site is: http://www.psycho.uni-duesseldorf.de/abteilungen/aap/gpower3/
That may get you through the power computation question
itself. My second comment is about this statement "As I can
estimate the statistical power to conclude that my results are valid." I'm
guessing that English is not your first language so there may be word choice
issues. A high power number is not going to establish 'validity'. A high power
number will show the degree of confidence that you can
have
that you would have found a
relationship of the specified size it it were really present. 'Validity' depends
on the correctness of your experimental design and statistical analysis logic.
Gene Maguin
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of MARGOT mh Sent: Tuesday, June 14, 2011 4:41 PM To: [hidden email] Subject: help me I have a database with 600 subjects who carried out, a
secondary statistical analysis with logistic regression with 5 independent
variables. As I can estimate the statistical power to conclude that my results
are valid.
Maga |
|
|
I might add to Gene's comments that large sample sizes yield high power (depending on the effect size in question). Assuming that the real population effect size values of your predictors are moderate
in magnitude (or larger) then you should have high power with that sample size.
Your question seems to be closer in meaning to a computation of the confidence intervals around your estimates. Large samples yield narrow confidence intervals, which one could think as having greater precision in the estimate
of the where the population parameters are. Again, this does not answer the "validity" question, for the same reasons Gene outlined.
Evan Harrington, Ph.D.
Department of Forensic Psychology
The Chicago School of Professional Psychology
325 N. Wells Street
Chicago, IL 60654
From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Gene Maguin [[hidden email]] Sent: Wednesday, June 15, 2011 9:22 AM To: [hidden email] Subject: Re: help me Maga,
It sounds like you did a logistic regression with 5 predictors (independent variables) using data from a sample with an N of 600. You now want to compute power to
show that your results are valid. I want to make two comments. One comment is about the power computation specifically. First of all, you can not do what you want using spss. You have to find something else.
Do a search on 'statistical power logistic regression' or a set of search terms like that. One link is to a program called G*Power 3.
I've never used this program, which seems to be free, but i think it could be a good option for you. The site is:
http://www.psycho.uni-duesseldorf.de/abteilungen/aap/gpower3/
That may get you through the power computation question itself. My second comment is about this statement "As I can estimate the statistical
power to conclude that my results are valid." I'm guessing that English is not your first language so there may be word choice issues. A high power number is not going to establish 'validity'. A high power number will show the degree of confidence that you
can have
that you would have found a relationship of the specified size it it were really present. 'Validity' depends on the correctness of your experimental
design and statistical analysis logic.
Gene Maguin
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of MARGOT mh Sent: Tuesday, June 14, 2011 4:41 PM To: [hidden email] Subject: help me I have a database with 600 subjects who carried out, a secondary statistical analysis with logistic regression with 5 independent variables. As I can estimate the statistical power to conclude that my results are valid.
Maga |
|
|
In reply to this post by Maguin, Eugene
Dear Gene and Evan,
Thank very much for your help me. I got a model (GEE) with 5 independent variables (2 and 3 were categories were continuous covariates), these are all significant. Not consider covariates did not influence my model. As this analysis is an analysis of a secondary database, I have requested that I submit the statistical power. My English is not very good, I hope I could make myself understood. My understanding is that the sample is large gives a high statistical power, so I do not understand is referred to the effect size. Thanks in advance. Maga > Date: Wed, 15 Jun 2011 11:45:37 -0500 > From: [hidden email] > Subject: Re: help me > To: [hidden email] > > --_000_5F28CD1102E6274BA7BE4796B8EA90F50A46047431TCSTCSESORG_ > Content-Type: text/plain; charset="iso-8859-1" > Content-Transfer-Encoding: quoted-printable > > I might add to Gene's comments that large sample sizes yield high power (de= > pending on the effect size in question). Assuming that the real population = > effect size values of your predictors are moderate in magnitude (or larger)= > then you should have high power with that sample size. > > Your question seems to be closer in meaning to a computation of the confide= > nce intervals around your estimates. Large samples yield narrow confidence = > intervals, which one could think as having greater precision in the estimat= > e of the where the population parameters are. Again, this does not answer t= > he "validity" question, for the same reasons Gene outlined. > > > Evan Harrington, Ph.D. > Department of Forensic Psychology > The Chicago School of Professional Psychology > 325 N. Wells Street > Chicago, IL 60654 > ________________________________ > From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Gene Magu= > in [[hidden email]] > Sent: Wednesday, June 15, 2011 9:22 AM > To: [hidden email] > Subject: Re: help me > > Maga, > > It sounds like you did a logistic regression with 5 predictors (independent= > variables) using data from a sample with an N of 600. You now want to comp= > ute power to show that your results are valid. I want to make two comments.= > One comment is about the power computation specifically. First of all, you= > can not do what you want using spss. You have to find something else. Do a= > search on 'statistical power logistic regression' or a set of search terms= > like that. One link is to a program called G*Power 3. I've never used this= > program, which seems to be free, but i think it could be a good option for= > you. The site is: http://www.psycho.uni-duesseldorf.de/abteilungen/aap/gpo= > wer3/ > > That may get you through the power computation question itself. My second c= > omment is about this statement "As I can estimate the statistical power to = > conclude that my results are valid." I'm guessing that English is not your = > first language so there may be word choice issues. A high power number is n= > ot going to establish 'validity'. A high power number will show the degree = > of confidence that you can have > that you would have found a relationship of the specified size it it were r= > eally present. 'Validity' depends on the correctness of your experimental d= > esign and statistical analysis logic. > > Gene Maguin > > > > ________________________________ > From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of MA= > RGOT mh > Sent: Tuesday, June 14, 2011 4:41 PM > To: [hidden email] > Subject: help me > > Dear all, > > I have a database with 600 subjects who carried out, a secondary statistica= > l analysis with logistic regression with 5 independent variables. As I can = > estimate the statistical power to conclude that my results are valid. > > Thank you very much for your answers > > Maga > > --_000_5F28CD1102E6274BA7BE4796B8EA90F50A46047431TCSTCSESORG_ > Content-Type: text/html; charset="iso-8859-1" > Content-Transfer-Encoding: quoted-printable > > <html dir=3D"ltr"><head> > <meta http-equiv=3D"Content-Type" content=3D"text/html; charset=3Diso-8859-= > 1"> > <style>.hmmessage P { > PADDING-RIGHT: 0px; PADDING-LEFT: 0px; PADDING-BOTTOM: 0px; MARGIN: 0px; P= > ADDING-TOP: 0px > } > BODY.hmmessage { > FONT-SIZE: 10pt; FONT-FAMILY: Tahoma > } > </style> > <meta content=3D"MSHTML 6.00.6000.17037" name=3D"GENERATOR"> > <style title=3D"owaParaStyle"><!--P { > MARGIN-TOP: 0px; MARGIN-BOTTOM: 0px > } > --></style> > </head> > <body class=3D"hmmessage" ocsi=3D"x"> > <div dir=3D"ltr"><font face=3D"Arial" color=3D"#000000" size=3D"2">I might = > add to Gene's comments that large sample sizes yield high power (depending = > on the effect size in question). Assuming that the real population effect s= > ize values of your predictors are moderate > in magnitude (or larger) then you should have high power with that sample = > size.</font></div> > <div dir=3D"ltr"><font face=3D"arial"></font> </div> > <div dir=3D"ltr"><font face=3D"arial">Your question seems to be closer in m= > eaning to a computation of the confidence intervals around your estimates. = > Large samples yield narrow confidence intervals, which one could think as h= > aving greater precision in the estimate > of the where the population parameters are. Again, this does not answer th= > e "validity" question, for the same reasons Gene outlined.</font>= > </div> > <div dir=3D"ltr"><font face=3D"arial"></font> </div> > <div dir=3D"ltr"><font face=3D"arial"></font> </div> > <div> > <div><font face=3D"Arial" size=3D"2">Evan Harrington, Ph.D.</font></div> > <div><font face=3D"arial" size=3D"2">Department of Forensic Psychology</fon= > t></div> > <div><font face=3D"arial" size=3D"2">The Chicago School of Professional Psy= > chology</font></div> > <div><font face=3D"arial" size=3D"2">325 N. Wells Street</font></div> > <div><font face=3D"arial" size=3D"2">Chicago, IL 60654</font></div> > </div> > <div id=3D"divRpF171723" style=3D"DIRECTION: ltr"> > <hr tabindex=3D"-1"> > <font face=3D"Tahoma" size=3D"2"><b>From:</b> SPSSX(r) Discussion [SPSSX-L@= > LISTSERV.UGA.EDU] On Behalf Of Gene Maguin [[hidden email]]<br> > <b>Sent:</b> Wednesday, June 15, 2011 9:22 AM<br> > <b>To:</b> [hidden email]<br> > <b>Subject:</b> Re: help me<br> > </font><br> > </div> > <div></div> > <div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"><spa= > n class=3D"314294013-15062011">Maga, > </span></font></div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"><spa= > n class=3D"314294013-15062011"></span></font> </div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"><spa= > n class=3D"314294013-15062011">It sounds like you did a logistic regression= > with 5 predictors (independent variables) using data from a sample with an= > N of 600. You now want to compute power to > show that your results are valid. I want to make two comments. One&nb= > sp;comment is about the power computation specifically. First of = > all, you can not do what you want using spss. You have to find something el= > se. > </span></font><font face=3D"Arial" color=3D"#0000ff"><span class=3D"3142940= > 13-15062011">Do a search on 'statistical power logistic regression' or a se= > t of search terms like that. One link is to a program called G*Power 3. > </span></font><font face=3D"Arial" color=3D"#0000ff"><span class=3D"3142940= > 13-15062011">I've never used this program, which seems to be free, but= > i think it could be a good option for you. The site is: > </span></font><font face=3D"Arial" color=3D"#0000ff"><span class=3D"3142940= > 13-15062011"><a href=3D"http://www.psycho.uni-duesseldorf.de/abteilungen/aa= > p/gpower3/" target=3D"_blank">http://www.psycho.uni-duesseldorf.de/abteilun= > gen/aap/gpower3/</a> > </span></font></div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"><spa= > n class=3D"314294013-15062011"></span></font> </div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"><spa= > n class=3D"314294013-15062011">That may get you through the power computati= > on question itself. My second comment is about this statement "<font c= > olor=3D"#000000">As I can estimate the statistical > power to conclude that my results are valid." I'm guessing that Engli= > sh is not your first language so there may be word choice issues. A high po= > wer number is not going to establish 'validity'. A high power number will s= > how the degree of confidence that you > can have</font></span></font></div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"><spa= > n class=3D"314294013-15062011"><font color=3D"#000000">that you would have = > found a relationship of the specified size it it were really present. 'Vali= > dity' depends on the correctness of your experimental > design and statistical analysis logic. </font></span></font></div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"><spa= > n class=3D"314294013-15062011"></span></font> </div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"><spa= > n class=3D"314294013-15062011">Gene Maguin</span></font></div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"></fo= > nt> </div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"></fo= > nt> </div> > <font face=3D"Arial" color=3D"#0000ff"></font><br> > <div class=3D"OutlookMessageHeader" lang=3D"en-us" dir=3D"ltr" align=3D"lef= > t"> > <hr tabindex=3D"-1"> > <font face=3D"Tahoma"><b>From:</b> SPSSX(r) Discussion [mailto:SPSSX-L@LIST= > SERV.UGA.EDU] > <b>On Behalf Of </b>MARGOT mh<br> > <b>Sent:</b> Tuesday, June 14, 2011 4:41 PM<br> > <b>To:</b> [hidden email]<br> > <b>Subject:</b> help me<br> > </font><br> > </div> > <div></div> > <font face=3D"Arial" size=3D"3">Dear all,</font><br> > <font face=3D"Arial" size=3D"3"></font> <br> > <p class=3D"MsoNormal" style=3D"MARGIN: 0cm 0cm 0pt; VERTICAL-ALIGN: top; L= > INE-HEIGHT: normal"> > <span lang=3D"EN" style=3D"COLOR: black; FONT-FAMILY: 'Arial','sans-serif'"= > ><font size=3D"3">I have a database with 600 subjects who carried out, a se= > condary statistical analysis with logistic regression with 5 independent va= > riables. As I can estimate the statistical > power to conclude that my results are valid.</font></span></p> > <p class=3D"MsoNormal" style=3D"MARGIN: 0cm 0cm 0pt; VERTICAL-ALIGN: top; L= > INE-HEIGHT: normal"> > <span lang=3D"EN" style=3D"COLOR: black; FONT-FAMILY: 'Arial','sans-serif'"= > ><br> > <font size=3D"3">Thank you very much for your answers</font></span><span la= > ng=3D"EN-GB" style=3D"FONT-SIZE: 9pt; COLOR: #888888; FONT-FAMILY: 'Arial',= > 'sans-serif'"></span></p> > <font face=3D"Arial" size=3D"3"></font> <br> > <font face=3D"Arial" size=3D"3">Maga</font><br> > </div> > </body> > </html> > > --_000_5F28CD1102E6274BA7BE4796B8EA90F50A46047431TCSTCSESORG_-- > > ===================== > To manage your subscription to SPSSX-L, send a message to > [hidden email] (not to SPSSX-L), with no body text except the > command. To leave the list, send the command > SIGNOFF SPSSX-L > For a list of commands to manage subscriptions, send the command > INFO REFCARD |
|
|
Maga,
Why are you using GEE? What is there about your
analysis or dataset that requires you to use GEE?
If you really have to use GEE, then the little program
i mentioned won't work and i don't know what type of software would be required
to compute power.
Gene Maguin From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of MARGOT mh Sent: Wednesday, June 15, 2011 5:01 PM To: [hidden email] Subject: Re: help me Thank very much for your help me. I got a model (GEE) with 5 independent variables (2 and 3 were categories were continuous covariates), these are all significant. Not consider covariates did not influence my model. As this analysis is an analysis of a secondary database, I have requested that I submit the statistical power. My English is not very good, I hope I could make myself understood. My understanding is that the sample is large gives a high statistical power, so I do not understand is referred to the effect size. Thanks in advance. Maga > Date: Wed, 15 Jun 2011 11:45:37 -0500 > From: [hidden email] > Subject: Re: help me > To: [hidden email] > > --_000_5F28CD1102E6274BA7BE4796B8EA90F50A46047431TCSTCSESORG_ > Content-Type: text/plain; charset="iso-8859-1" > Content-Transfer-Encoding: quoted-printable > > I might add to Gene's comments that large sample sizes yield high power (de= > pending on the effect size in question). Assuming that the real population = > effect size values of your predictors are moderate in magnitude (or larger)= > then you should have high power with that sample size. > > Your question seems to be closer in meaning to a computation of the confide= > nce intervals around your estimates. Large samples yield narrow confidence = > intervals, which one could think as having greater precision in the estimat= > e of the where the population parameters are. Again, this does not answer t= > he "validity" question, for the same reasons Gene outlined. > > > Evan Harrington, Ph.D. > Department of Forensic Psychology > The Chicago School of Professional Psychology > 325 N. Wells Street > Chicago, IL 60654 > ________________________________ > From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Gene Magu= > in [[hidden email]] > Sent: Wednesday, June 15, 2011 9:22 AM > To: [hidden email] > Subject: Re: help me > > Maga, > > It sounds like you did a logistic regression with 5 predictors (independent= > variables) using data from a sample with an N of 600. You now want to comp= > ute power to show that your results are valid. I want to make two comments.= > One comment is about the power computation specifically. First of all, you= > can not do what you want using spss. You have to find something else. Do a= > search on 'statistical power logistic regression' or a set of search terms= > like that. One link is to a program called G*Power 3. I've never used this= > program, which seems to be free, but i think it could be a good option for= > you. The site is: http://www.psycho.uni-duesseldorf.de/abteilungen/aap/gpo= > wer3/ > > That may get you through the power computation question itself. My second c= > omment is about this statement "As I can estimate the statistical power to = > conclude that my results are valid." I'm guessing that English is not your = > first language so there may be word choice issues. A high power number is n= > ot going to establish 'validity'. A high power number will show the degree = > of confidence that you can have > that you would have found a relationship of the specified size it it were r= > eally present. 'Validity' depends on the correctness of your experimental d= > esign and statistical analysis logic. > > Gene Maguin > > > > ________________________________ > From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of MA= > RGOT mh > Sent: Tuesday, June 14, 2011 4:41 PM > To: [hidden email] > Subject: help me > > Dear all, > > I have a database with 600 subjects who carried out, a secondary statistica= > l analysis with logistic regression with 5 independent variables. As I can = > estimate the statistical power to conclude that my results are valid. > > Thank you very much for your answers > > Maga > > --_000_5F28CD1102E6274BA7BE4796B8EA90F50A46047431TCSTCSESORG_ > Content-Type: text/html; charset="iso-8859-1" > Content-Transfer-Encoding: quoted-printable > > <html dir=3D"ltr"><head> > <meta http-equiv=3D"Content-Type" content=3D"text/html; charset=3Diso-8859-= > 1"> > <style>.hmmessage P { > PADDING-RIGHT: 0px; PADDING-LEFT: 0px; PADDING-BOTTOM: 0px; MARGIN: 0px; P= > ADDING-TOP: 0px > } > BODY.hmmessage { > FONT-SIZE: 10pt; FONT-FAMILY: Tahoma > } > </style> > <meta content=3D"MSHTML 6.00.6000.17037" name=3D"GENERATOR"> > <style title=3D"owaParaStyle"><!--P { > MARGIN-TOP: 0px; MARGIN-BOTTOM: 0px > } > --></style> > </head> > <body class=3D"hmmessage" ocsi=3D"x"> > <div dir=3D"ltr"><font face=3D"Arial" color=3D"#000000" size=3D"2">I might = > add to Gene's comments that large sample sizes yield high power (depending = > on the effect size in question). Assuming that the real population effect s= > ize values of your predictors are moderate > in magnitude (or larger) then you should have high power with that sample = > size.</font></div> > <div dir=3D"ltr"><font face=3D"arial"></font> </div> > <div dir=3D"ltr"><font face=3D"arial">Your question seems to be closer in m= > eaning to a computation of the confidence intervals around your estimates. = > Large samples yield narrow confidence intervals, which one could think as h= > aving greater precision in the estimate > of the where the population parameters are. Again, this does not answer th= > e "validity" question, for the same reasons Gene outlined.</font>= > </div> > <div dir=3D"ltr"><font face=3D"arial"></font> </div> > <div dir=3D"ltr"><font face=3D"arial"></font> </div> > <div> > <div><font face=3D"Arial" size=3D"2">Evan Harrington, Ph.D.</font></div> > <div><font face=3D"arial" size=3D"2">Department of Forensic Psychology</fon= > t></div> > <div><font face=3D"arial" size=3D"2">The Chicago School of Professional Psy= > chology</font></div> > <div><font face=3D"arial" size=3D"2">325 N. Wells Street</font></div> > <div><font face=3D"arial" size=3D"2">Chicago, IL 60654</font></div> > </div> > <div id=3D"divRpF171723" style=3D"DIRECTION: ltr"> > <hr tabindex=3D"-1"> > <font face=3D"Tahoma" size=3D"2"><b>From:</b> SPSSX(r) Discussion [SPSSX-L@= > LISTSERV.UGA.EDU] On Behalf Of Gene Maguin [[hidden email]]<br> > <b>Sent:</b> Wednesday, June 15, 2011 9:22 AM<br> > <b>To:</b> [hidden email]<br> > <b>Subject:</b> Re: help me<br> > </font><br> > </div> > <div></div> > <div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"><spa= > n class=3D"314294013-15062011">Maga, > </span></font></div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"><spa= > n class=3D"314294013-15062011"></span></font> </div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"><spa= > n class=3D"314294013-15062011">It sounds like you did a logistic regression= > with 5 predictors (independent variables) using data from a sample with an= > N of 600. You now want to compute power to > show that your results are valid. I want to make two comments. One&nb= > sp;comment is about the power computation specifically. First of = > all, you can not do what you want using spss. You have to find something el= > se. > </span></font><font face=3D"Arial" color=3D"#0000ff"><span class=3D"3142940= > 13-15062011">Do a search on 'statistical power logistic regression' or a se= > t of search terms like that. One link is to a program called G*Power 3. > </span></font><font face=3D"Arial" color=3D"#0000ff"><span class=3D"3142940= > 13-15062011">I've never used this program, which seems to be free, but= > i think it could be a good option for you. The site is: > </span></font><font face=3D"Arial" color=3D"#0000ff"><span class=3D"3142940= > 13-15062011"><a href=3D"http://www.psycho.uni-duesseldorf.de/abteilungen/aa= > p/gpower3/" target=3D"_blank">http://www.psycho.uni-duesseldorf.de/abteilun= > gen/aap/gpower3/</a> > </span></font></div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"><spa= > n class=3D"314294013-15062011"></span></font> </div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"><spa= > n class=3D"314294013-15062011">That may get you through the power computati= > on question itself. My second comment is about this statement "<font c= > olor=3D"#000000">As I can estimate the statistical > power to conclude that my results are valid." I'm guessing that Engli= > sh is not your first language so there may be word choice issues. A high po= > wer number is not going to establish 'validity'. A high power number will s= > how the degree of confidence that you > can have</font></span></font></div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"><spa= > n class=3D"314294013-15062011"><font color=3D"#000000">that you would have = > found a relationship of the specified size it it were really present. 'Vali= > dity' depends on the correctness of your experimental > design and statistical analysis logic. </font></span></font></div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"><spa= > n class=3D"314294013-15062011"></span></font> </div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"><spa= > n class=3D"314294013-15062011">Gene Maguin</span></font></div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"></fo= > nt> </div> > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" color=3D"#0000ff"></fo= > nt> </div> > <font face=3D"Arial" color=3D"#0000ff"></font><br> > <div class=3D"OutlookMessageHeader" lang=3D"en-us" dir=3D"ltr" align=3D"lef= > t"> > <hr tabindex=3D"-1"> > <font face=3D"Tahoma"><b>From:</b> SPSSX(r) Discussion [mailto:SPSSX-L@LIST= > SERV.UGA.EDU] > <b>On Behalf Of </b>MARGOT mh<br> > <b>Sent:</b> Tuesday, June 14, 2011 4:41 PM<br> > <b>To:</b> [hidden email]<br> > <b>Subject:</b> help me<br> > </font><br> > </div> > <div></div> > <font face=3D"Arial" size=3D"3">Dear all,</font><br> > <font face=3D"Arial" size=3D"3"></font> <br> > <p class=3D"MsoNormal" style=3D"MARGIN: 0cm 0cm 0pt; VERTICAL-ALIGN: top; L= > INE-HEIGHT: normal"> > <span lang=3D"EN" style=3D"COLOR: black; FONT-FAMILY: 'Arial','sans-serif'"= > ><font size=3D"3">I have a database with 600 subjects who carried out, a se= > condary statistical analysis with logistic regression with 5 independent va= > riables. As I can estimate the statistical > power to conclude that my results are valid.</font></span></p> > <p class=3D"MsoNormal" style=3D"MARGIN: 0cm 0cm 0pt; VERTICAL-ALIGN: top; L= > INE-HEIGHT: normal"> > <span lang=3D"EN" style=3D"COLOR: black; FONT-FAMILY: 'Arial','sans-serif'"= > ><br> > <font size=3D"3">Thank you very much for your answers</font></span><span la= > ng=3D"EN-GB" style=3D"FONT-SIZE: 9pt; COLOR: #888888; FONT-FAMILY: 'Arial',= > 'sans-serif'"></span></p> > <font face=3D"Arial" size=3D"3"></font> <br> > <font face=3D"Arial" size=3D"3">Maga</font><br> > </div> > </body> > </html> > > --_000_5F28CD1102E6274BA7BE4796B8EA90F50A46047431TCSTCSESORG_-- > > ===================== > To manage your subscription to SPSSX-L, send a message to > [hidden email] (not to SPSSX-L), with no body text except the > command. To leave the list, send the command > SIGNOFF SPSSX-L > For a list of commands to manage subscriptions, send the command > INFO REFCARD |
|
|
In reply to this post by MARGOT mh
You clearly had *enough* statistical power, given the significant
findings! Instead of focusing on post-hoc statistical power, I'd be much more interested in examining the odds ratios and respective confidence intervals. On Wed, Jun 15, 2011 at 5:00 PM, MARGOT mh <[hidden email]> wrote: > Dear Gene and Evan, > > Thank very much for your help me. > > I got a model (GEE) with 5 independent variables (2 and 3 were categories > were continuous covariates), these are all significant. Not consider > covariates did not influence my model. As this analysis is an analysis of a > secondary database, I have requested that I submit the statistical power. My > English is not very good, I hope I could make myself understood. > My understanding is that the sample is large gives a high statistical power, > so I do not understand is referred to the effect size. > > Thanks in advance. > > Maga > > > >> Date: Wed, 15 Jun 2011 11:45:37 -0500 >> From: [hidden email] >> Subject: Re: help me >> To: [hidden email] >> >> --_000_5F28CD1102E6274BA7BE4796B8EA90F50A46047431TCSTCSESORG_ >> Content-Type: text/plain; charset="iso-8859-1" >> Content-Transfer-Encoding: quoted-printable >> >> I might add to Gene's comments that large sample sizes yield high power >> (de= >> pending on the effect size in question). Assuming that the real population >> = >> effect size values of your predictors are moderate in magnitude (or >> larger)= >> then you should have high power with that sample size. >> >> Your question seems to be closer in meaning to a computation of the >> confide= >> nce intervals around your estimates. Large samples yield narrow confidence >> = >> intervals, which one could think as having greater precision in the >> estimat= >> e of the where the population parameters are. Again, this does not answer >> t= >> he "validity" question, for the same reasons Gene outlined. >> >> >> Evan Harrington, Ph.D. >> Department of Forensic Psychology >> The Chicago School of Professional Psychology >> 325 N. Wells Street >> Chicago, IL 60654 >> ________________________________ >> From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Gene >> Magu= >> in [[hidden email]] >> Sent: Wednesday, June 15, 2011 9:22 AM >> To: [hidden email] >> Subject: Re: help me >> >> Maga, >> >> It sounds like you did a logistic regression with 5 predictors >> (independent= >> variables) using data from a sample with an N of 600. You now want to >> comp= >> ute power to show that your results are valid. I want to make two >> comments.= >> One comment is about the power computation specifically. First of all, >> you= >> can not do what you want using spss. You have to find something else. Do >> a= >> search on 'statistical power logistic regression' or a set of search >> terms= >> like that. One link is to a program called G*Power 3. I've never used >> this= >> program, which seems to be free, but i think it could be a good option >> for= >> you. The site is: >> http://www.psycho.uni-duesseldorf.de/abteilungen/aap/gpo= >> wer3/ >> >> That may get you through the power computation question itself. My second >> c= >> omment is about this statement "As I can estimate the statistical power to >> = >> conclude that my results are valid." I'm guessing that English is not your >> = >> first language so there may be word choice issues. A high power number is >> n= >> ot going to establish 'validity'. A high power number will show the degree >> = >> of confidence that you can have >> that you would have found a relationship of the specified size it it were >> r= >> eally present. 'Validity' depends on the correctness of your experimental >> d= >> esign and statistical analysis logic. >> >> Gene Maguin >> >> >> >> ________________________________ >> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of >> MA= >> RGOT mh >> Sent: Tuesday, June 14, 2011 4:41 PM >> To: [hidden email] >> Subject: help me >> >> Dear all, >> >> I have a database with 600 subjects who carried out, a secondary >> statistica= >> l analysis with logistic regression with 5 independent variables. As I can >> = >> estimate the statistical power to conclude that my results are valid. >> >> Thank you very much for your answers >> >> Maga >> >> --_000_5F28CD1102E6274BA7BE4796B8EA90F50A46047431TCSTCSESORG_ >> Content-Type: text/html; charset="iso-8859-1" >> Content-Transfer-Encoding: quoted-printable >> >> <html dir=3D"ltr"><head> >> <meta http-equiv=3D"Content-Type" content=3D"text/html; >> charset=3Diso-8859-= >> 1"> >> <style>.hmmessage P { >> PADDING-RIGHT: 0px; PADDING-LEFT: 0px; PADDING-BOTTOM: 0px; MARGIN: 0px; >> P= >> ADDING-TOP: 0px >> } >> BODY.hmmessage { >> FONT-SIZE: 10pt; FONT-FAMILY: Tahoma >> } >> </style> >> <meta content=3D"MSHTML 6.00.6000.17037" name=3D"GENERATOR"> >> <style title=3D"owaParaStyle"><!--P { >> MARGIN-TOP: 0px; MARGIN-BOTTOM: 0px >> } >> --></style> >> </head> >> <body class=3D"hmmessage" ocsi=3D"x"> >> <div dir=3D"ltr"><font face=3D"Arial" color=3D"#000000" size=3D"2">I might >> = >> add to Gene's comments that large sample sizes yield high power (depending >> = >> on the effect size in question). Assuming that the real population effect >> s= >> ize values of your predictors are moderate >> in magnitude (or larger) then you should have high power with that sample >> = >> size.</font></div> >> <div dir=3D"ltr"><font face=3D"arial"></font> </div> >> <div dir=3D"ltr"><font face=3D"arial">Your question seems to be closer in >> m= >> eaning to a computation of the confidence intervals around your estimates. >> = >> Large samples yield narrow confidence intervals, which one could think as >> h= >> aving greater precision in the estimate >> of the where the population parameters are. Again, this does not answer >> th= >> e "validity" question, for the same reasons Gene >> outlined.</font>= >> </div> >> <div dir=3D"ltr"><font face=3D"arial"></font> </div> >> <div dir=3D"ltr"><font face=3D"arial"></font> </div> >> <div> >> <div><font face=3D"Arial" size=3D"2">Evan Harrington, Ph.D.</font></div> >> <div><font face=3D"arial" size=3D"2">Department of Forensic >> Psychology</fon= >> t></div> >> <div><font face=3D"arial" size=3D"2">The Chicago School of Professional >> Psy= >> chology</font></div> >> <div><font face=3D"arial" size=3D"2">325 N. Wells Street</font></div> >> <div><font face=3D"arial" size=3D"2">Chicago, IL 60654</font></div> >> </div> >> <div id=3D"divRpF171723" style=3D"DIRECTION: ltr"> >> <hr tabindex=3D"-1"> >> <font face=3D"Tahoma" size=3D"2"><b>From:</b> SPSSX(r) Discussion >> [SPSSX-L@= >> LISTSERV.UGA.EDU] On Behalf Of Gene Maguin [[hidden email]]<br> >> <b>Sent:</b> Wednesday, June 15, 2011 9:22 AM<br> >> <b>To:</b> [hidden email]<br> >> <b>Subject:</b> Re: help me<br> >> </font><br> >> </div> >> <div></div> >> <div> >> <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" >> color=3D"#0000ff"><spa= >> n class=3D"314294013-15062011">Maga, >> </span></font></div> >> <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" >> color=3D"#0000ff"><spa= >> n class=3D"314294013-15062011"></span></font> </div> >> <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" >> color=3D"#0000ff"><spa= >> n class=3D"314294013-15062011">It sounds like you did a logistic >> regression= >> with 5 predictors (independent variables) using data from a sample with >> an= >> N of 600. You now want to compute power to >> show that your results are valid. I want to make two comments. >> One&nb= >> sp;comment is about the power computation specifically. First of >> = >> all, you can not do what you want using spss. You have to find something >> el= >> se. >> </span></font><font face=3D"Arial" color=3D"#0000ff"><span >> class=3D"3142940= >> 13-15062011">Do a search on 'statistical power logistic regression' or a >> se= >> t of search terms like that. One link is to a program called G*Power 3. >> </span></font><font face=3D"Arial" color=3D"#0000ff"><span >> class=3D"3142940= >> 13-15062011">I've never used this program, which seems to be >> free, but= >> i think it could be a good option for you. The site is: >> </span></font><font face=3D"Arial" color=3D"#0000ff"><span >> class=3D"3142940= >> 13-15062011"><a >> href=3D"http://www.psycho.uni-duesseldorf.de/abteilungen/aa= >> p/gpower3/" >> target=3D"_blank">http://www.psycho.uni-duesseldorf.de/abteilun= >> gen/aap/gpower3/</a> >> </span></font></div> >> <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" >> color=3D"#0000ff"><spa= >> n class=3D"314294013-15062011"></span></font> </div> >> <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" >> color=3D"#0000ff"><spa= >> n class=3D"314294013-15062011">That may get you through the power >> computati= >> on question itself. My second comment is about this statement "<font >> c= >> olor=3D"#000000">As I can estimate the statistical >> power to conclude that my results are valid." I'm guessing that >> Engli= >> sh is not your first language so there may be word choice issues. A high >> po= >> wer number is not going to establish 'validity'. A high power number will >> s= >> how the degree of confidence that you >> can have</font></span></font></div> >> <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" >> color=3D"#0000ff"><spa= >> n class=3D"314294013-15062011"><font color=3D"#000000">that you would have >> = >> found a relationship of the specified size it it were really present. >> 'Vali= >> dity' depends on the correctness of your experimental >> design and statistical analysis logic. </font></span></font></div> >> <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" >> color=3D"#0000ff"><spa= >> n class=3D"314294013-15062011"></span></font> </div> >> <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" >> color=3D"#0000ff"><spa= >> n class=3D"314294013-15062011">Gene Maguin</span></font></div> >> <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" >> color=3D"#0000ff"></fo= >> nt> </div> >> <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" >> color=3D"#0000ff"></fo= >> nt> </div> >> <font face=3D"Arial" color=3D"#0000ff"></font><br> >> <div class=3D"OutlookMessageHeader" lang=3D"en-us" dir=3D"ltr" >> align=3D"lef= >> t"> >> <hr tabindex=3D"-1"> >> <font face=3D"Tahoma"><b>From:</b> SPSSX(r) Discussion >> [mailto:SPSSX-L@LIST= >> SERV.UGA.EDU] >> <b>On Behalf Of </b>MARGOT mh<br> >> <b>Sent:</b> Tuesday, June 14, 2011 4:41 PM<br> >> <b>To:</b> [hidden email]<br> >> <b>Subject:</b> help me<br> >> </font><br> >> </div> >> <div></div> >> <font face=3D"Arial" size=3D"3">Dear all,</font><br> >> <font face=3D"Arial" size=3D"3"></font> <br> >> <p class=3D"MsoNormal" style=3D"MARGIN: 0cm 0cm 0pt; VERTICAL-ALIGN: top; >> L= >> INE-HEIGHT: normal"> >> <span lang=3D"EN" style=3D"COLOR: black; FONT-FAMILY: >> 'Arial','sans-serif'"= >> ><font size=3D"3">I have a database with 600 subjects who carried out, a >> > se= >> condary statistical analysis with logistic regression with 5 independent >> va= >> riables. As I can estimate the statistical >> power to conclude that my results are valid.</font></span></p> >> <p class=3D"MsoNormal" style=3D"MARGIN: 0cm 0cm 0pt; VERTICAL-ALIGN: top; >> L= >> INE-HEIGHT: normal"> >> <span lang=3D"EN" style=3D"COLOR: black; FONT-FAMILY: >> 'Arial','sans-serif'"= >> ><br> >> <font size=3D"3">Thank you very much for your answers</font></span><span >> la= >> ng=3D"EN-GB" style=3D"FONT-SIZE: 9pt; COLOR: #888888; FONT-FAMILY: >> 'Arial',= >> 'sans-serif'"></span></p> >> <font face=3D"Arial" size=3D"3"></font> <br> >> <font face=3D"Arial" size=3D"3">Maga</font><br> >> </div> >> </body> >> </html> >> >> --_000_5F28CD1102E6274BA7BE4796B8EA90F50A46047431TCSTCSESORG_-- >> >> ===================== >> To manage your subscription to SPSSX-L, send a message to >> [hidden email] (not to SPSSX-L), with no body text except the >> command. To leave the list, send the command >> SIGNOFF SPSSX-L >> For a list of commands to manage subscriptions, send the command >> INFO REFCARD > ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
|
|
In reply to this post by MARGOT mh
Hi Margot,
No, G power won't work. I don't have enough experience
to offer any more advice. I don't understand the necessity of a post-analysis
power computation. I saw that Ryan Black posted a reply and i think
his suggestions are very much worth considering. However, if a power computation
is absolutely required you will have to find somebody where you are to do the
computation for you.
Sorry, Gene Maguin
From: MARGOT mh
[mailto:[hidden email]]
Sent: Thursday, June 16, 2011 9:54 AM To: [hidden email]; [hidden email] Subject: RE: help me Dear Gene, I have got 3 repeated measurements per subject (correlated observations), and while they are in clusters. I have got a longitudinal study with 3 measurements are also subject groups. Then I don´t use G power???... Just thank you very much for your feedback. Maga > Date: Wed, 15 Jun 2011 17:16:25 -0400 > From: [hidden email] > Subject: Re: help me > To: [hidden email] > > This is a multi-part message in MIME format. > > ------=_NextPart_000_002D_01CC2B7F.F4FA7F40 > Content-Type: text/plain; > charset="us-ascii" > Content-Transfer-Encoding: 7bit > > Maga, > > Why are you using GEE? What is there about your analysis or dataset that > requires you to use GEE? > > If you really have to use GEE, then the little program i mentioned won't > work and i don't know what type of software would be required to compute > power. > > Gene Maguin > > _____ > > From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of > MARGOT mh > Sent: Wednesday, June 15, 2011 5:01 PM > To: [hidden email] > Subject: Re: help me > > > Dear Gene and Evan, > > Thank very much for your help me. > > I got a model (GEE) with 5 independent variables (2 and 3 were categories > were continuous covariates), these are all significant. Not consider > covariates did not influence my model. As this analysis is an analysis of a > secondary database, I have requested that I submit the statistical power. My > English is not very good, I hope I could make myself understood. > My understanding is that the sample is large gives a high statistical power, > so I do not understand is referred to the effect size. > > Thanks in advance. > > Maga > > > > > Date: Wed, 15 Jun 2011 11:45:37 -0500 > > From: [hidden email] > > Subject: Re: help me > > To: [hidden email] > > > > --_000_5F28CD1102E6274BA7BE4796B8EA90F50A46047431TCSTCSESORG_ > > Content-Type: text/plain; charset="iso-8859-1" > > Content-Transfer-Encoding: quoted-printable > > > > I might add to Gene's comments that large sample sizes yield high power > (de= > > pending on the effect size in question). Assuming that the real population > = > > effect size values of your predictors are moderate in magnitude (or > larger)= > > then you should have high power with that sample size. > > > > Your question seems to be closer in meaning to a computation of the > confide= > > nce intervals around your estimates. Large samples yield narrow confidence > = > > intervals, which one could think as having greater precision in the > estimat= > > e of the where the population parameters are. Again, this does not answer > t= > > he "validity" question, for the same reasons Gene outlined. > > > > > > Evan Harrington, Ph.D. > > Department of Forensic Psychology > > The Chicago School of Professional Psychology > > 325 N. Wells Street > > Chicago, IL 60654 > > ________________________________ > > From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Gene > Magu= > > in [[hidden email]] > > Sent: Wednesday, June 15, 2011 9:22 AM > > To: [hidden email] > > Subject: Re: help me > > > > Maga, > > > > It sounds like you did a logistic regression with 5 predictors > (independent= > > variables) using data from a sample with an N of 600. You now want to > comp= > > ute power to show that your results are valid. I want to make two > comments.= > > One comment is about the power computation specifically. First of all, > you= > > can not do what you want using spss. You have to find something else. Do > a= > > search on 'statistical power logistic regression' or a set of search > terms= > > like that. One link is to a program called G*Power 3. I've never used > this= > > program, which seems to be free, but i think it could be a good option > for= > > you. The site is: > http://www.psycho.uni-duesseldorf.de/abteilungen/aap/gpo= > > wer3/ > > > > That may get you through the power computation question itself. My second > c= > > omment is about this statement "As I can estimate the statistical power to > = > > conclude that my results are valid." I'm guessing that English is not your > = > > first language so there may be word choice issues. A high power number is > n= > > ot going to establish 'validity'. A high power number will show the degree > = > > of confidence that you can have > > that you would have found a relationship of the specified size it it were > r= > > eally present. 'Validity' depends on the correctness of your experimental > d= > > esign and statistical analysis logic. > > > > Gene Maguin > > > > > > > > ________________________________ > > From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of > MA= > > RGOT mh > > Sent: Tuesday, June 14, 2011 4:41 PM > > To: [hidden email] > > Subject: help me > > > > Dear all, > > > > I have a database with 600 subjects who carried out, a secondary > statistica= > > l analysis with logistic regression with 5 independent variables. As I can > = > > estimate the statistical power to conclude that my results are valid. > > > > Thank you very much for your answers > > > > Maga > > > > --_000_5F28CD1102E6274BA7BE4796B8EA90F50A46047431TCSTCSESORG_ > > Content-Type: text/html; charset="iso-8859-1" > > Content-Transfer-Encoding: quoted-printable > > > > <html dir=3D"ltr"><head> > > <meta http-equiv=3D"Content-Type" content=3D"text/html; > charset=3Diso-8859-= > > 1"> > > <style>.hmmessage P { > > PADDING-RIGHT: 0px; PADDING-LEFT: 0px; PADDING-BOTTOM: 0px; MARGIN: 0px; > P= > > ADDING-TOP: 0px > > } > > BODY.hmmessage { > > FONT-SIZE: 10pt; FONT-FAMILY: Tahoma > > } > > </style> > > <meta content=3D"MSHTML 6.00.6000.17037" name=3D"GENERATOR"> > > <style title=3D"owaParaStyle"><!--P { > > MARGIN-TOP: 0px; MARGIN-BOTTOM: 0px > > } > > --></style> > > </head> > > <body class=3D"hmmessage" ocsi=3D"x"> > > <div dir=3D"ltr"><font face=3D"Arial" color=3D"#000000" size=3D"2">I might > = > > add to Gene's comments that large sample sizes yield high power (depending > = > > on the effect size in question). Assuming that the real population effect > s= > > ize values of your predictors are moderate > > in magnitude (or larger) then you should have high power with that sample > = > > size.</font></div> > > <div dir=3D"ltr"><font face=3D"arial"></font> </div> > > <div dir=3D"ltr"><font face=3D"arial">Your question seems to be closer in > m= > > eaning to a computation of the confidence intervals around your estimates. > = > > Large samples yield narrow confidence intervals, which one could think as > h= > > aving greater precision in the estimate > > of the where the population parameters are. Again, this does not answer > th= > > e "validity" question, for the same reasons Gene > outlined.</font>= > > </div> > > <div dir=3D"ltr"><font face=3D"arial"></font> </div> > > <div dir=3D"ltr"><font face=3D"arial"></font> </div> > > <div> > > <div><font face=3D"Arial" size=3D"2">Evan Harrington, Ph.D.</font></div> > > <div><font face=3D"arial" size=3D"2">Department of Forensic > Psychology</fon= > > t></div> > > <div><font face=3D"arial" size=3D"2">The Chicago School of Professional > Psy= > > chology</font></div> > > <div><font face=3D"arial" size=3D"2">325 N. Wells Street</font></div> > > <div><font face=3D"arial" size=3D"2">Chicago, IL 60654</font></div> > > </div> > > <div id=3D"divRpF171723" style=3D"DIRECTION: ltr"> > > <hr tabindex=3D"-1"> > > <font face=3D"Tahoma" size=3D"2"><b>From:</b> SPSSX(r) Discussion > [SPSSX-L@= > > LISTSERV.UGA.EDU] On Behalf Of Gene Maguin [[hidden email]]<br> > > <b>Sent:</b> Wednesday, June 15, 2011 9:22 AM<br> > > <b>To:</b> [hidden email]<br> > > <b>Subject:</b> Re: help me<br> > > </font><br> > > </div> > > <div></div> > > <div> > > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" > color=3D"#0000ff"><spa= > > n class=3D"314294013-15062011">Maga, > > </span></font></div> > > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" > color=3D"#0000ff"><spa= > > n class=3D"314294013-15062011"></span></font> </div> > > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" > color=3D"#0000ff"><spa= > > n class=3D"314294013-15062011">It sounds like you did a logistic > regression= > > with 5 predictors (independent variables) using data from a sample with > an= > > N of 600. You now want to compute power to > > show that your results are valid. I want to make two comments. > One&nb= > > sp;comment is about the power computation specifically. First of > = > > all, you can not do what you want using spss. You have to find something > el= > > se. > > </span></font><font face=3D"Arial" color=3D"#0000ff"><span > class=3D"3142940= > > 13-15062011">Do a search on 'statistical power logistic regression' or a > se= > > t of search terms like that. One link is to a program called G*Power 3. > > </span></font><font face=3D"Arial" color=3D"#0000ff"><span > class=3D"3142940= > > 13-15062011">I've never used this program, which seems to be > free, but= > > i think it could be a good option for you. The site is: > > </span></font><font face=3D"Arial" color=3D"#0000ff"><span > class=3D"3142940= > > 13-15062011"><a > href=3D"http://www.psycho.uni-duesseldorf.de/abteilungen/aa= > > p/gpower3/" > target=3D"_blank">http://www.psycho.uni-duesseldorf.de/abteilun= > > gen/aap/gpower3/</a> > > </span></font></div> > > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" > color=3D"#0000ff"><spa= > > n class=3D"314294013-15062011"></span></font> </div> > > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" > color=3D"#0000ff"><spa= > > n class=3D"314294013-15062011">That may get you through the power > computati= > > on question itself. My second comment is about this statement "<font > c= > > olor=3D"#000000">As I can estimate the statistical > > power to conclude that my results are valid." I'm guessing that > Engli= > > sh is not your first language so there may be word choice issues. A high > po= > > wer number is not going to establish 'validity'. A high power number will > s= > > how the degree of confidence that you > > can have</font></span></font></div> > > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" > color=3D"#0000ff"><spa= > > n class=3D"314294013-15062011"><font color=3D"#000000">that you would have > = > > found a relationship of the specified size it it were really present. > 'Vali= > > dity' depends on the correctness of your experimental > > design and statistical analysis logic. </font></span></font></div> > > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" > color=3D"#0000ff"><spa= > > n class=3D"314294013-15062011"></span></font> </div> > > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" > color=3D"#0000ff"><spa= > > n class=3D"314294013-15062011">Gene Maguin</span></font></div> > > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" > color=3D"#0000ff"></fo= > > nt> </div> > > <div dir=3D"ltr" align=3D"left"><font face=3D"Arial" > color=3D"#0000ff"></fo= > > nt> </div> > > <font face=3D"Arial" color=3D"#0000ff"></font><br> > > <div class=3D"OutlookMessageHeader" lang=3D"en-us" dir=3D"ltr" > align=3D"lef= > > t"> > > <hr tabindex=3D"-1"> > > <font face=3D"Tahoma"><b>From:</b> SPSSX(r) Discussion > [mailto:SPSSX-L@LIST= > > SERV.UGA.EDU] > > <b>On Behalf Of </b>MARGOT mh<br> > > <b>Sent:</b> Tuesday, June 14, 2011 4:41 PM<br> > > <b>To:</b> [hidden email]<br> > > <b>Subject:</b> help me<br> > > </font><br> > > </div> > > <div></div> > > <font face=3D"Arial" size=3D"3">Dear all,</font><br> > > <font face=3D"Arial" size=3D"3"></font> <br> > > <p class=3D"MsoNormal" style=3D"MARGIN: 0cm 0cm 0pt; VERTICAL-ALIGN: top; > L= > > INE-HEIGHT: normal"> > > <span lang=3D"EN" style=3D"COLOR: black; FONT-FAMILY: > 'Arial','sans-serif'"= > > ><font size=3D"3">I have a database with 600 subjects who carried out, a > se= > > condary statistical analysis with logistic regression with 5 independent > va= > > riables. As I can estimate the statistical > > power to conclude that my results are valid.</font></span></p> > > <p class=3D"MsoNormal" style=3D"MARGIN: 0cm 0cm 0pt; VERTICAL-ALIGN: top; > L= > > INE-HEIGHT: normal"> > > <span lang=3D"EN" style=3D"COLOR: black; FONT-FAMILY: > 'Arial','sans-serif'"= > > ><br> > > <font size=3D"3">Thank you very much for your answers</font></span><span > la= > > ng=3D"EN-GB" style=3D"FONT-SIZE: 9pt; COLOR: #888888; FONT-FAMILY: > 'Arial',= > > 'sans-serif'"></span></p> > > <font face=3D"Arial" size=3D"3"></font> <br> > > <font face=3D"Arial" size=3D"3">Maga</font><br> > > </div> > > </body> > > </html> > > > > --_000_5F28CD1102E6274BA7BE4796B8EA90F50A46047431TCSTCSESORG_-- > > > > ===================== > > To manage your subscription to SPSSX-L, send a message to > > [hidden email] (not to SPSSX-L), with no body text except the > > command. To leave the list, send the command > > SIGNOFF SPSSX-L > > For a list of commands to manage subscriptions, send the command > > INFO REFCARD > > > ------=_NextPart_000_002D_01CC2B7F.F4FA7F40 > Content-Type: text/html; > charset="us-ascii" > Content-Transfer-Encoding: quoted-printable > > <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN"> > <HTML><HEAD> > <META content=3D"text/html; charset=3Dus-ascii" = > http-equiv=3DContent-Type> > <STYLE>.hmmessage P { > PADDING-BOTTOM: 0px; MARGIN: 0px; PADDING-LEFT: 0px; PADDING-RIGHT: = > 0px; PADDING-TOP: 0px > } > BODY.hmmessage { > FONT-FAMILY: Tahoma; FONT-SIZE: 10pt > } > </STYLE> > > <META name=3DGENERATOR content=3D"MSHTML 8.00.6001.19088"></HEAD> > <BODY class=3Dhmmessage> > <DIV dir=3Dltr align=3Dleft><FONT color=3D#0000ff face=3DArial><SPAN=20 > class=3D604241121-15062011>Maga, </SPAN></FONT></DIV> > <DIV dir=3Dltr align=3Dleft><FONT color=3D#0000ff face=3DArial><SPAN=20 > class=3D604241121-15062011></SPAN></FONT> </DIV> > <DIV dir=3Dltr align=3Dleft><FONT color=3D#0000ff face=3DArial><SPAN=20 > class=3D604241121-15062011>Why are you using GEE? What is there = > about your=20 > analysis or dataset that requires you to use GEE?</SPAN></FONT></DIV> > <DIV dir=3Dltr align=3Dleft><FONT color=3D#0000ff face=3DArial><SPAN=20 > class=3D604241121-15062011></SPAN></FONT> </DIV> > <DIV dir=3Dltr align=3Dleft><FONT color=3D#0000ff face=3DArial><SPAN=20 > class=3D604241121-15062011>If you really have to use GEE, then the = > little program=20 > i mentioned won't work and i don't know what type of software would be = > required=20 > to compute power. </SPAN></FONT></DIV> > <DIV dir=3Dltr align=3Dleft><FONT color=3D#0000ff face=3DArial><SPAN=20 > class=3D604241121-15062011></SPAN></FONT> </DIV> > <DIV dir=3Dltr align=3Dleft><FONT color=3D#0000ff face=3DArial><SPAN=20 > class=3D604241121-15062011>Gene Maguin</SPAN></FONT></DIV><BR> > <DIV dir=3Dltr lang=3Den-us class=3DOutlookMessageHeader align=3Dleft> > <HR tabIndex=3D-1> > <FONT face=3DTahoma><B>From:</B> SPSSX(r) Discussion=20 > [mailto:[hidden email]] <B>On Behalf Of </B>MARGOT = > mh<BR><B>Sent:</B>=20 > Wednesday, June 15, 2011 5:01 PM<BR><B>To:</B>=20 > [hidden email]<BR><B>Subject:</B> Re: help = > me<BR></FONT><BR></DIV> > <DIV></DIV>Dear Gene and Evan,<BR> <BR>Thank very much for your = > help me.=20 > <BR> <BR><SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1079">I</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1080">got</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1081">a model</SPAN> <SPAN class=3D"hps atn"=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1082">(</SPAN><SPAN=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1083">GEE</SPAN><SPAN=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1084">) with</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1085">5</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1086">independent</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1087">variables</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1088">(2</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1089">and</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1090">3</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1091">were</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1092">categories</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1093">were</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1094">continuous covariates</SPAN><SPAN=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1095">)</SPAN><SPAN=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1096">,</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1097">these are all</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1098">significant</SPAN><SPAN=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1099">.</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1100">Not</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1101">consider</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1102">covariates</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1103">did not influence</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1104">my model</SPAN><SPAN=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1105">.</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1106">As</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1107">this</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1108">analysis</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1109">is</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1110">an</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1111">analysis of</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1112">a</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1113">secondary database</SPAN><SPAN=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1114">,</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1115">I</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1116">have requested</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1117">that</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1118">I</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1119">submit</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1120">the statistical power</SPAN><SPAN=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1121">.</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1122">My</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1123">English</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1124">is not</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1125">very</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1126">good</SPAN><SPAN=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1127">,</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1128">I hope I</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1129">could</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1130">make myself understood.</SPAN><BR=20 > closure_uid_gabdcu=3D"1151"><SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1131">My understanding</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1132">is</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1133">that</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1134">the</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1135">sample</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1136">is</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1137">large</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1138">gives</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1139">a</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1140">high</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1141">statistical power</SPAN><SPAN=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1142">,</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1143">so</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1144">I do not understand</SPAN> <SPAN class=3Dhps = > > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1145">is</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1146">referred to</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1147">the</SPAN> <SPAN class=3Dhps=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1148">effect size</SPAN><SPAN=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1149">.</SPAN><BR><SPAN=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1149"></SPAN> <BR><SPAN=20 > title=3D"Haz clic para obtener otras posibles traducciones"=20 > closure_uid_gabdcu=3D"1149">Thanks in advance.<BR><BR>Maga</SPAN><BR><BR = > > closure_uid_gabdcu=3D"1152"><BR> <BR>> Date: Wed, 15 Jun 2011 = > 11:45:37=20 > -0500<BR>> From: [hidden email]<BR>> Subject: = > Re: help=20 > me<BR>> To: [hidden email]<BR>> <BR>>=20 > --_000_5F28CD1102E6274BA7BE4796B8EA90F50A46047431TCSTCSESORG_<BR>>=20 > Content-Type: text/plain; charset=3D"iso-8859-1"<BR>>=20 > Content-Transfer-Encoding: quoted-printable<BR>> <BR>> I might add = > to=20 > Gene's comments that large sample sizes yield high power (de=3D<BR>> = > pending on=20 > the effect size in question). Assuming that the real population = > =3D<BR>> effect=20 > size values of your predictors are moderate in magnitude (or = > larger)=3D<BR>>=20 > then you should have high power with that sample size.<BR>> <BR>> = > Your=20 > question seems to be closer in meaning to a computation of the = > confide=3D<BR>>=20 > nce intervals around your estimates. Large samples yield narrow = > confidence=20 > =3D<BR>> intervals, which one could think as having greater precision = > in the=20 > estimat=3D<BR>> e of the where the population parameters are. Again, = > this does=20 > not answer t=3D<BR>> he "validity" question, for the same reasons = > Gene=20 > outlined.<BR>> <BR>> <BR>> Evan Harrington, Ph.D.<BR>> = > Department of=20 > Forensic Psychology<BR>> The Chicago School of Professional=20 > Psychology<BR>> 325 N. Wells Street<BR>> Chicago, IL 60654<BR>> = > > ________________________________<BR>> From: SPSSX(r) Discussion=20 > [[hidden email]] On Behalf Of Gene Magu=3D<BR>> in=20 > [[hidden email]]<BR>> Sent: Wednesday, June 15, 2011 9:22 = > AM<BR>> To:=20 > [hidden email]<BR>> Subject: Re: help me<BR>> <BR>>=20 > Maga,<BR>> <BR>> It sounds like you did a logistic regression with = > 5=20 > predictors (independent=3D<BR>> variables) using data from a sample = > with an N=20 > of 600. You now want to comp=3D<BR>> ute power to show that your = > results are=20 > valid. I want to make two comments.=3D<BR>> One comment is about the = > power=20 > computation specifically. First of all, you=3D<BR>> can not do what = > you want=20 > using spss. You have to find something else. Do a=3D<BR>> search on=20 > 'statistical power logistic regression' or a set of search = > terms=3D<BR>> like=20 > that. One link is to a program called G*Power 3. I've never used = > this=3D<BR>>=20 > program, which seems to be free, but i think it could be a good option=20 > for=3D<BR>> you. The site is:=20 > http://www.psycho.uni-duesseldorf.de/abteilungen/aap/gpo=3D<BR>> = > wer3/<BR>>=20 > <BR>> That may get you through the power computation question itself. = > My=20 > second c=3D<BR>> omment is about this statement "As I can estimate = > the=20 > statistical power to =3D<BR>> conclude that my results are valid." = > I'm guessing=20 > that English is not your =3D<BR>> first language so there may be word = > choice=20 > issues. A high power number is n=3D<BR>> ot going to establish = > 'validity'. A=20 > high power number will show the degree =3D<BR>> of confidence that = > you can=20 > have<BR>> that you would have found a relationship of the specified = > size it=20 > it were r=3D<BR>> eally present. 'Validity' depends on the = > correctness of your=20 > experimental d=3D<BR>> esign and statistical analysis logic.<BR>> = > <BR>>=20 > Gene Maguin<BR>> <BR>> <BR>> <BR>>=20 > ________________________________<BR>> From: SPSSX(r) Discussion=20 > [mailto:[hidden email]] On Behalf Of MA=3D<BR>> RGOT = > mh<BR>> Sent:=20 > Tuesday, June 14, 2011 4:41 PM<BR>> To: = > [hidden email]<BR>>=20 > Subject: help me<BR>> <BR>> Dear all,<BR>> <BR>> I have a = > database=20 > with 600 subjects who carried out, a secondary statistica=3D<BR>> l = > analysis=20 > with logistic regression with 5 independent variables. As I can = > =3D<BR>>=20 > estimate the statistical power to conclude that my results are = > valid.<BR>>=20 > <BR>> Thank you very much for your answers<BR>> <BR>> = > Maga<BR>>=20 > <BR>> = > --_000_5F28CD1102E6274BA7BE4796B8EA90F50A46047431TCSTCSESORG_<BR>>=20 > Content-Type: text/html; charset=3D"iso-8859-1"<BR>> = > Content-Transfer-Encoding:=20 > quoted-printable<BR>> <BR>> <html = > dir=3D3D"ltr"><head><BR>>=20 > <meta http-equiv=3D3D"Content-Type" content=3D3D"text/html;=20 > charset=3D3Diso-8859-=3D<BR>> 1"><BR>> <style>.hmmessage = > P {<BR>>=20 > PADDING-RIGHT: 0px; PADDING-LEFT: 0px; PADDING-BOTTOM: 0px; MARGIN: 0px; = > > P=3D<BR>> ADDING-TOP: 0px<BR>> }<BR>> BODY.hmmessage {<BR>> = > FONT-SIZE:=20 > 10pt; FONT-FAMILY: Tahoma<BR>> }<BR>> </style><BR>> = > <meta=20 > content=3D3D"MSHTML 6.00.6000.17037" name=3D3D"GENERATOR"><BR>> = > <style=20 > title=3D3D"owaParaStyle"><!--P {<BR>> MARGIN-TOP: 0px; = > MARGIN-BOTTOM:=20 > 0px<BR>> }<BR>> --></style><BR>> </head><BR>> = > > <body class=3D3D"hmmessage" ocsi=3D3D"x"><BR>> <div=20 > dir=3D3D"ltr"><font face=3D3D"Arial" color=3D3D"#000000" = > size=3D3D"2">I might=20 > =3D<BR>> add to Gene's comments that large sample sizes yield high = > power=20 > (depending =3D<BR>> on the effect size in question). Assuming that = > the real=20 > population effect s=3D<BR>> ize values of your predictors are = > moderate<BR>>=20 > in magnitude (or larger) then you should have high power with that = > sample=20 > =3D<BR>> size.</font></div><BR>> <div = > dir=3D3D"ltr"><font=20 > face=3D3D"arial"></font>&nbsp;</div><BR>> <div=20 > dir=3D3D"ltr"><font face=3D3D"arial">Your question seems to be = > closer in=20 > m=3D<BR>> eaning to a computation of the confidence intervals around = > your=20 > estimates. =3D<BR>> Large samples yield narrow confidence intervals, = > which one=20 > could think as h=3D<BR>> aving greater precision in the = > estimate<BR>> of the=20 > where the population parameters are. Again, this does not answer = > th=3D<BR>> e=20 > &quot;validity&quot; question, for the same reasons Gene=20 > outlined.</font>=3D<BR>> </div><BR>> <div=20 > dir=3D3D"ltr"><font=20 > face=3D3D"arial"></font>&nbsp;</div><BR>> <div=20 > dir=3D3D"ltr"><font=20 > face=3D3D"arial"></font>&nbsp;</div><BR>>=20 > <div><BR>> <div><font face=3D3D"Arial" = > size=3D3D"2">Evan=20 > Harrington, Ph.D.</font></div><BR>> <div><font=20 > face=3D3D"arial" size=3D3D"2">Department of Forensic = > Psychology</fon=3D<BR>>=20 > t></div><BR>> <div><font face=3D3D"arial" = > size=3D3D"2">The=20 > Chicago School of Professional Psy=3D<BR>>=20 > chology</font></div><BR>> <div><font = > face=3D3D"arial"=20 > size=3D3D"2">325 N. Wells Street</font></div><BR>>=20 > <div><font face=3D3D"arial" size=3D3D"2">Chicago, IL=20 > 60654</font></div><BR>> </div><BR>> <div=20 > id=3D3D"divRpF171723" style=3D3D"DIRECTION: ltr"><BR>> <hr=20 > tabindex=3D3D"-1"><BR>> <font face=3D3D"Tahoma"=20 > size=3D3D"2"><b>From:</b> SPSSX(r) Discussion = > [SPSSX-L@=3D<BR>>=20 > LISTSERV.UGA.EDU] On Behalf Of Gene Maguin=20 > [[hidden email]]<br><BR>> <b>Sent:</b> = > Wednesday, June=20 > 15, 2011 9:22 AM<br><BR>> <b>To:</b>=20 > [hidden email]<br><BR>> <b>Subject:</b> = > Re: help=20 > me<br><BR>> </font><br><BR>> = > </div><BR>>=20 > <div></div><BR>> <div><BR>> <div = > dir=3D3D"ltr"=20 > align=3D3D"left"><font face=3D3D"Arial" = > color=3D3D"#0000ff"><spa=3D<BR>>=20 > n class=3D3D"314294013-15062011">Maga,<BR>>=20 > </span></font></div><BR>> <div dir=3D3D"ltr"=20 > align=3D3D"left"><font face=3D3D"Arial" = > color=3D3D"#0000ff"><spa=3D<BR>>=20 > n=20 > class=3D3D"314294013-15062011"></span></font>&nbsp;<= > ;/div><BR>>=20 > <div dir=3D3D"ltr" align=3D3D"left"><font face=3D3D"Arial"=20 > color=3D3D"#0000ff"><spa=3D<BR>> n = > class=3D3D"314294013-15062011">It=20 > sounds like you did a logistic regression=3D<BR>> with 5 predictors=20 > (independent variables) using data from a sample with an=3D<BR>> N of = > 600. You=20 > now want to compute power&nbsp;to<BR>> show that your results are = > > valid.&nbsp;I want to make two comments. One&nb=3D<BR>> = > sp;comment=20 > is&nbsp;about the power computation specifically.&nbsp;First of=20 > =3D<BR>> all, you can not do what you want using spss. You have to = > find=20 > something el=3D<BR>> se.<BR>> </span></font><font=20 > face=3D3D"Arial" color=3D3D"#0000ff"><span = > class=3D3D"3142940=3D<BR>>=20 > 13-15062011">Do a search on 'statistical power logistic regression' = > or a=20 > se=3D<BR>> t of search terms like that. One link is to a program = > called G*Power=20 > 3.<BR>> </span></font><font face=3D3D"Arial"=20 > color=3D3D"#0000ff"><span class=3D3D"3142940=3D<BR>> = > 13-15062011">I've=20 > never used this program, which seems to be free,&nbsp;but=3D<BR>> = > i think=20 > it could be a good option for you. The site is:<BR>>=20 > </span></font><font face=3D3D"Arial" = > color=3D3D"#0000ff"><span=20 > class=3D3D"3142940=3D<BR>> 13-15062011"><a=20 > href=3D3D"http://www.psycho.uni-duesseldorf.de/abteilungen/aa=3D<BR>> = > p/gpower3/"=20 > target=3D3D"_blank">http://www.psycho.uni-duesseldorf.de/abteilun=3D<B= > R>>=20 > gen/aap/gpower3/</a>&nbsp;<BR>>=20 > </span></font></div><BR>> <div dir=3D3D"ltr"=20 > align=3D3D"left"><font face=3D3D"Arial" = > color=3D3D"#0000ff"><spa=3D<BR>>=20 > n=20 > class=3D3D"314294013-15062011"></span></font>&nbsp;<= > ;/div><BR>>=20 > <div dir=3D3D"ltr" align=3D3D"left"><font face=3D3D"Arial"=20 > color=3D3D"#0000ff"><spa=3D<BR>> n = > class=3D3D"314294013-15062011">That may=20 > get you through the power computati=3D<BR>> on question itself. My = > second=20 > comment is about this statement &quot;<font c=3D<BR>>=20 > olor=3D3D"#000000">As I can estimate the statistical<BR>> power to = > conclude=20 > that my results are valid.&quot; I'm guessing that Engli=3D<BR>> = > sh is not=20 > your first language so there may be word choice issues. A high = > po=3D<BR>> wer=20 > number is not going to establish 'validity'. A high power number will = > s=3D<BR>>=20 > how the degree of confidence that you<BR>> can=20 > have</font></span></font></div><BR>> <div=20 > dir=3D3D"ltr" align=3D3D"left"><font face=3D3D"Arial"=20 > color=3D3D"#0000ff"><spa=3D<BR>> n = > class=3D3D"314294013-15062011"><font=20 > color=3D3D"#000000">that you would have =3D<BR>> found a = > relationship of the=20 > specified size it it were really present. 'Vali=3D<BR>> dity' depends = > on the=20 > correctness of your experimental<BR>> design and statistical analysis = > logic.=20 > </font></span></font></div><BR>> <div = > dir=3D3D"ltr"=20 > align=3D3D"left"><font face=3D3D"Arial" = > color=3D3D"#0000ff"><spa=3D<BR>>=20 > n=20 > class=3D3D"314294013-15062011"></span></font>&nbsp;<= > ;/div><BR>>=20 > <div dir=3D3D"ltr" align=3D3D"left"><font face=3D3D"Arial"=20 > color=3D3D"#0000ff"><spa=3D<BR>> n = > class=3D3D"314294013-15062011">Gene=20 > Maguin</span></font></div><BR>> <div = > dir=3D3D"ltr"=20 > align=3D3D"left"><font face=3D3D"Arial" = > color=3D3D"#0000ff"></fo=3D<BR>>=20 > nt>&nbsp;</div><BR>> <div dir=3D3D"ltr"=20 > align=3D3D"left"><font face=3D3D"Arial" = > color=3D3D"#0000ff"></fo=3D<BR>>=20 > nt>&nbsp;</div><BR>> <font face=3D3D"Arial"=20 > color=3D3D"#0000ff"></font><br><BR>> <div=20 > class=3D3D"OutlookMessageHeader" lang=3D3D"en-us" dir=3D3D"ltr" = > align=3D3D"lef=3D<BR>>=20 > t"><BR>> <hr tabindex=3D3D"-1"><BR>> <font=20 > face=3D3D"Tahoma"><b>From:</b> SPSSX(r) Discussion=20 > [mailto:SPSSX-L@LIST=3D<BR>> SERV.UGA.EDU]<BR>> <b>On Behalf = > Of=20 > </b>MARGOT mh<br><BR>> <b>Sent:</b> Tuesday, = > June 14,=20 > 2011 4:41 PM<br><BR>> <b>To:</b>=20 > [hidden email]<br><BR>> <b>Subject:</b> = > help=20 > me<br><BR>> </font><br><BR>> = > </div><BR>>=20 > <div></div><BR>> <font face=3D3D"Arial" = > size=3D3D"3">Dear=20 > all,</font><br><BR>> <font face=3D3D"Arial"=20 > size=3D3D"3"></font>&nbsp;<br><BR>> <p=20 > class=3D3D"MsoNormal" style=3D3D"MARGIN: 0cm 0cm 0pt; VERTICAL-ALIGN: = > top;=20 > L=3D<BR>> INE-HEIGHT: normal"><BR>> <span lang=3D3D"EN" = > style=3D3D"COLOR:=20 > black; FONT-FAMILY: 'Arial','sans-serif'"=3D<BR>> ><font = > size=3D3D"3">I=20 > have a database with 600 subjects who carried out, a se=3D<BR>> = > condary=20 > statistical analysis with logistic regression with 5 independent = > va=3D<BR>>=20 > riables. As I can estimate the statistical<BR>> power to conclude = > that my=20 > results are valid.</font></span></p><BR>> <p=20 > class=3D3D"MsoNormal" style=3D3D"MARGIN: 0cm 0cm 0pt; VERTICAL-ALIGN: = > top;=20 > L=3D<BR>> INE-HEIGHT: normal"><BR>> <span lang=3D3D"EN" = > style=3D3D"COLOR:=20 > black; FONT-FAMILY: 'Arial','sans-serif'"=3D<BR>> = > ><br><BR>>=20 > <font size=3D3D"3">Thank you very much for your=20 > answers</font></span><span la=3D<BR>> ng=3D3D"EN-GB"=20 > style=3D3D"FONT-SIZE: 9pt; COLOR: #888888; FONT-FAMILY: = > 'Arial',=3D<BR>>=20 > 'sans-serif'"></span></p><BR>> <font = > face=3D3D"Arial"=20 > size=3D3D"3"></font>&nbsp;<br><BR>> <font = > face=3D3D"Arial"=20 > size=3D3D"3">Maga</font><br><BR>> </div><BR>> = > > </body><BR>> </html><BR>> <BR>>=20 > --_000_5F28CD1102E6274BA7BE4796B8EA90F50A46047431TCSTCSESORG_--<BR>> = > <BR>>=20 > =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D<BR>> = > To manage your subscription to SPSSX-L, send a=20 > message to<BR>> [hidden email] (not to SPSSX-L), with no = > body text=20 > except the<BR>> command. To leave the list, send the command<BR>> = > SIGNOFF=20 > SPSSX-L<BR>> For a list of commands to manage subscriptions, send the = > > command<BR>> INFO REFCARD<BR></BODY></HTML> > > ------=_NextPart_000_002D_01CC2B7F.F4FA7F40-- > > ===================== > To manage your subscription to SPSSX-L, send a message to > [hidden email] (not to SPSSX-L), with no body text except the > command. To leave the list, send the command > SIGNOFF SPSSX-L > For a list of commands to manage subscriptions, send the command > INFO REFCARD |
|
|
In reply to this post by MARGOT mh
The usual proper use of a power analysis is to describe, even before
data collection, what the chances are that your tests will be able to detect ("find significant") effect sizes that are (a) interesting, and (b) at least somewhat likely to be seen, given the range of Ns that is being considered. In your instance, it *might* be reasonable that someone could be asking that you show, for your sample, how small of an effect could have been detected, say, with even 50% power. If some of your tests were not significant, it could be interesting to remark about how large of an effect could have been missed. (I also consider it possible that the person who asked for your power analysis is an idiot, who is mindlessly echoing, inappropriately, jargon that he has heard in the past. It possible to make some descriptive statement, after the analysis, referring to power, but I agree with the poster who says that giving confidence limits is often going to be more informative, to more readers.) A sample size of 600 is "large" for what Cohen was interested in, for the usual studies in psychology and social sciences, and their usual effects. However, there are other topics where 600 is far too small. The study which showed that aspirin after a heart-attack is a good idea was the first "mega-study" using multiple sites, and it was designed from the power analysis which showed that 20,000 patients was a good number, if one wanted to show that a treatment that would cut in half the immediate fatalities was worthwhile. What is an "effect size"? For a dichotomy like a dummy variable, predicting another dichotomy, the odds-ratio is a good measure of effect. However, the Odds Ratio is sensitive to the actual sizes of the proportions, so you often need to anchor the description in the actual proportions. That is, there is a 20-point difference between 40% and 60%, or between 10% and 30%; but the latter is a more severe difference than the former. For your study, your purpose is some sort of illustration -- Obviously, you had *enough* power for some analyses, since you did have positive findings. I think I would perform the power analyses one variable at a time. That makes the presentation easier and clearer, and it is often the approach to start with, even the power analysis is done before the experiment. -- Rich Ulrich Date: Wed, 15 Jun 2011 21:00:31 +0000 From: [hidden email] Subject: Re: help me To: [hidden email] Dear Gene and Evan, Thank very much for your help me. I got a model (GEE) with 5 independent variables (2 and 3 were categories were continuous covariates), these are all significant. Not consider covariates did not influence my model. As this analysis is an analysis of a secondary database, I have requested that I submit the statistical power. My English is not very good, I hope I could make myself understood. My understanding is that the sample is large gives a high statistical power, so I do not understand is referred to the effect size. Thanks in advance. Maga [snip, previous] |
|
|
Rich, thank you very much. your comments have been invaluable to me and I have cleared many doubts.
regards Maga From: [hidden email] To: [hidden email]; [hidden email] Subject: RE: help me Date: Thu, 16 Jun 2011 14:16:30 -0400 The usual proper use of a power analysis is to describe, even before data collection, what the chances are that your tests will be able to detect ("find significant") effect sizes that are (a) interesting, and (b) at least somewhat likely to be seen, given the range of Ns that is being considered. In your instance, it *might* be reasonable that someone could be asking that you show, for your sample, how small of an effect could have been detected, say, with even 50% power. If some of your tests were not significant, it could be interesting to remark about how large of an effect could have been missed. (I also consider it possible that the person who asked for your power analysis is an idiot, who is mindlessly echoing, inappropriately, jargon that he has heard in the past. It possible to make some descriptive statement, after the analysis, referring to power, but I agree with the poster who says that giving confidence limits is often going to be more informative, to more readers.) A sample size of 600 is "large" for what Cohen was interested in, for the usual studies in psychology and social sciences, and their usual effects. However, there are other topics where 600 is far too small. The study which showed that aspirin after a heart-attack is a good idea was the first "mega-study" using multiple sites, and it was designed from the power analysis which showed that 20,000 patients was a good number, if one wanted to show that a treatment that would cut in half the immediate fatalities was worthwhile. What is an "effect size"? For a dichotomy like a dummy variable, predicting another dichotomy, the odds-ratio is a good measure of effect. However, the Odds Ratio is sensitive to the actual sizes of the proportions, so you often need to anchor the description in the actual proportions. That is, there is a 20-point difference between 40% and 60%, or between 10% and 30%; but the latter is a more severe difference than the former. For your study, your purpose is some sort of illustration -- Obviously, you had *enough* power for some analyses, since you did have positive findings. I think I would perform the power analyses one variable at a time. That makes the presentation easier and clearer, and it is often the approach to start with, even the power analysis is done before the experiment. -- Rich Ulrich Date: Wed, 15 Jun 2011 21:00:31 +0000 From: [hidden email] Subject: Re: help me To: [hidden email] Dear Gene and Evan, Thank very much for your help me. I got a model (GEE) with 5 independent variables (2 and 3 were categories were continuous covariates), these are all significant. Not consider covariates did not influence my model. As this analysis is an analysis of a secondary database, I have requested that I submit the statistical power. My English is not very good, I hope I could make myself understood. My understanding is that the sample is large gives a high statistical power, so I do not understand is referred to the effect size. Thanks in advance. Maga [snip, previous] |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |