MethodS of multivariate regression

[Kevin Smith]

Hi , I am analyzing medical patient cases with multivariate regression
modeling.



With SPSS, logistic regression, we can choose the method. I use method
ENTER, the Sig=0.03. If I choose stepwise (forward conditional Sig=0.07,
forward LR Sig=0.07, forward Wald Sig=0.07, backward conditional
Sig=0.03,  backward
LR Sig=0.03, backwald Wald Sig=0.03).



How do I interpret the significance of these results? My understanding is
that the method ENTER is the best. Never do stepwise. Please help.



Kevin

=====================
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

[Kevin Cai]

Could you please let me know the way to  calculate minimal sample size?

  Thank you.

  DRK


---------------------------------
Never miss a thing.   Make Yahoo your homepage.

=====================
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

[Hal 9000]

You need to determine power (the probability of rejecting the null
hypothesis when it's really false)..that depends on effect size,
significance level, and sample size. Decide what level of power you're
willing to live with - you'll probably want at least a 70% to 80%
chance of finding an effect - if it exists. That determines sample
size.

Be more specific and I'm sure you'll receive more replies.
-Gary

On Jan 26, 2008 7:05 AM, Kevin Cai <[hidden email]> wrote:

> Could you please let me know the way to  calculate minimal sample size?
>
>   Thank you.
>
>   DRK
>
>
> ---------------------------------
> Never miss a thing.   Make Yahoo your homepage.
>
> =====================
> 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

[Hal 9000]

don't forget to check spsstools.net:

http://spsstools.net/SampleSyntax.htm#SampleSize

=====================
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

[Sara House]

DRK -

You're talking about performing power analysis, and since you're wanting to figure out minimum sample size, I assume you're wanting to do a priori power analysis (what sample size do I need?) as opposed to a post hoc power analysis (how much power did I actually have?).  This type of analysis is based on two things:

1) how much power you actually want, which Gary did an excellent job of summarizing in his email

AND

2) the effect size associated with what you're studying.  This value could be found in past research, your best bet being a meta-analysis, which would give you an overall effect size for the phenomenon you're studying.

I believe there are programs that will conduct this power analysis for you once you input these values.  I've never actually performed an a priori power analysis (just post hoc with SPSS), so someone else will have to fill you in on those details.  Depending on what type of design your using and how you're going to analyze your data, there are rules of thumb for sample size, but a power analysis is definitely more precise.

Sara House
[hidden email]

=====================
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

[Nika Yugay]

Hey DRK,

You can do these calculations with G*Power. Just type in Googld ''G*Power'',
the link will pop up, and you download the latest version of the program.
There, you set a-priori (power analysis) and select the statistical
procedure you're employing. Then plug in the criteria pointed out earlier
(effect size and the desired level of power). Then press calculate, and the
minimum sample sample size will be calculated.

Good luck!

Nika

On Jan 28, 2008 12:27 AM, Sara House <[hidden email]> wrote:

> DRK -
>
> You're talking about performing power analysis, and since you're wanting
> to figure out minimum sample size, I assume you're wanting to do a priori
> power analysis (what sample size do I need?) as opposed to a post hoc power
> analysis (how much power did I actually have?).  This type of analysis is
> based on two things:
>
> 1) how much power you actually want, which Gary did an excellent job of
> summarizing in his email
>
> AND
>
> 2) the effect size associated with what you're studying.  This value could
> be found in past research, your best bet being a meta-analysis, which would
> give you an overall effect size for the phenomenon you're studying.
>
> I believe there are programs that will conduct this power analysis for you
> once you input these values.  I've never actually performed an a priori
> power analysis (just post hoc with SPSS), so someone else will have to fill
> you in on those details.  Depending on what type of design your using and
> how you're going to analyze your data, there are rules of thumb for sample
> size, but a power analysis is definitely more precise.
>
> Sara House
> [hidden email]
>
> =====================
> 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

[SoS Statistical Services]

I find Minitab easier to use than GPower for sample size/power calculations. Sara, how do you do post-hoc power analysis using SPSS?

  Thanks

  Evie

Sara House <[hidden email]> wrote:
  DRK -

You're talking about performing power analysis, and since you're wanting to figure out minimum sample size, I assume you're wanting to do a priori power analysis (what sample size do I need?) as opposed to a post hoc power analysis (how much power did I actually have?). This type of analysis is based on two things:

1) how much power you actually want, which Gary did an excellent job of summarizing in his email

AND

2) the effect size associated with what you're studying. This value could be found in past research, your best bet being a meta-analysis, which would give you an overall effect size for the phenomenon you're studying.

I believe there are programs that will conduct this power analysis for you once you input these values. I've never actually performed an a priori power analysis (just post hoc with SPSS), so someone else will have to fill you in on those details. Depending on what type of design your using and how you're going to analyze your data, there are rules of thumb for sample size, but a power analysis is definitely more precise.

Sara House
[hidden email]

=====================
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



Evie Gardner
SoS Statistical Services
Email: [hidden email]
Tel:   [phone] Mobile [phone]

---------------------------------
 Yahoo! Answers - Get better answers from someone who knows. Tryit now.

=====================
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

[SR Millis-3]

Post hoc power analysis is not recommended. See:

Hoenig, J. M., & Heisey, D. M. (2001). The abuse of
power: The pervasive fallacy of power calculations for
data analysis The American Statistician, 55(1), 19-24.

SR Millis

--- SoS Statistical Services <[hidden email]>
wrote:

> I find Minitab easier to use than GPower for sample
> size/power calculations. Sara, how do you do
> post-hoc power analysis using SPSS?
>
>   Thanks
>


Scott R Millis, PhD, MEd, ABPP (CN,CL,RP), CStat
Professor & Director of Research
Dept of Physical Medicine & Rehabilitation
Wayne State University School of Medicine
261 Mack Blvd
Detroit, MI 48201
Email:  [hidden email]
Tel: [phone]
Fax: [phone]

=====================
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

[Sara House]

Thanks, Scott, for the reference.  Looks like an interesting article - I'm looking forward to reading it.

I can definitely understand the authors' arguments about why post-hoc is a bad thing which is why I commend people who do a priori power analysis.  To answer a question posed earlier, I have used what the authors of this article (and SPSS) refer to as "observed power".  I wouldn't say it is the preferred method of determining power, since you're using the current experiment's values to estimate parameters to determine power - using theoretical values to determine theoretical power.  It seems that your observed power would be biased to be low if you found no effect, which could lead researchers to believe they failed to find an effect due to small sample size.  Perhaps their observed power was very high and there was actually no effect to be found.  But then again, isn't power completely theoretical anyway?  Can we ever truly know power?  I would argue that there are good ways and bad ways of estimating power, but you can never be absolutely sure.

And I agree that the fact that some journals and textbooks urge researchers to find out how much power they had in an experiment is alarming.  This is why in my research methods class, I do teach my students about power and explain to them the logic of determining power ahead of time.

Sara House
[hidden email]

>>> SR Millis <[hidden email]> 01/28/08 10:19 AM >>>
Post hoc power analysis is not recommended. See:

Hoenig, J. M., & Heisey, D. M. (2001). The abuse of
power: The pervasive fallacy of power calculations for
data analysis The American Statistician, 55(1), 19-24.

SR Millis
Scott R Millis, PhD, MEd, ABPP (CN,CL,RP), CStat
Professor & Director of Research
Dept of Physical Medicine & Rehabilitation
Wayne State University School of Medicine
261 Mack Blvd
Detroit, MI 48201
Email:  [hidden email]
Tel: [phone]
Fax: [phone]

=====================
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