Bootstrapping in SPSS 19

[chutson]

I have conducted an RCT looking at an intervention for dementia.  I have two groups - (control and experimental) and two sets of measures (pre and post).  I did want to perform an ANCOVA on the pre/post difference scores for the two groups but some of the data is not normal. I was therefore going to try bootstrapping:

Analyze > Descriptive Statistics > Explore...
Explore main dialog
Select difference score as a dependent variable.
Select Statistics in the Display group.
Click Bootstrap.
Select Perform bootstrapping.
select Set seed for Mersenne Twister and type
592004 as the seed.
To obtain more accurate intervals (at the cost of more processing time), select Bias corrected
accelerated (BCa)
Click Continue.
Click OK in the Explore dialog box.


But then bootstrapping doesnt come up in the output - instead it says: some or all bootstrap sample results are issing, so no bootstrap estimation has been performed for this table.

Can anyone help?
Thanks
Charlotte

[Valerie Villella]

Please note that I am out of the office on Friday June 8, 2012 and Monday June 11, 2012 at an OANHSS education workshop. I will return to the office on Tuesday June 12, 2012.

Thank you.

Valerie Villella
Education Coordinator and
Policy & Program Analyst
ext. 228

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

[Poes, Matthew Joseph]

I can't directly help with this but would strongly recommend you reconsider the pre and post difference scores in favor of an ANCOVA covariance for change model in which pre-tests are the covariate and post tests are the response variable.  It's a less biased approach.  Change scores are problematic and make a lot of untenable assumptions.

Matthew J Poes
Research Data Specialist
Center for Prevention Research and Development
University of Illinois
510 Devonshire Dr.
Champaign, IL 61820
Phone: [phone]
email: [hidden email]



-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of chutson
Sent: Friday, June 08, 2012 11:57 AM
To: [hidden email]
Subject: Bootstrapping in SPSS 19

I have conducted an RCT looking at an intervention for dementia.  I have two groups - (control and experimental) and two sets of measures (pre and post).
I did want to perform an ANCOVA on the pre/post difference scores for the two groups but some of the data is not normal. I was therefore going to try
bootstrapping:

Analyze > Descriptive Statistics > Explore...
Explore main dialog
Select difference score as a dependent variable.
Select Statistics in the Display group.
Click Bootstrap.
Select Perform bootstrapping.
select Set seed for Mersenne Twister and type
592004 as the seed.
To obtain more accurate intervals (at the cost of more processing time), select Bias corrected accelerated (BCa) Click Continue.
Click OK in the Explore dialog box.


But then bootstrapping doesnt come up in the output - instead it says: some or all bootstrap sample results are issing, so no bootstrap estimation has been performed for this table.

Can anyone help?
Thanks
Charlotte

--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Bootstrapping-in-SPSS-19-tp5713597.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

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

[Art Kendall]

How did you decide something was not normal?
Imagine a graphic with difference score on the horizontal axis and Frequency on the vertical axis.
One would want (hope, hypothesize) that to be bi-modal. 
In these General Linear Models of which ANCOVA is a special instance, one worries whether the _residuals_ are flagrantly disparate from normal.

Clearly 2 variables are quite non-normal. The groups that cases were randomized to is described by a dichotomous variable treatment vs control and time is described by pre vs post.  However these are on the independent/predictor/right-hand side.

What syntax did you use to do the original analysis?

What syntax did your use of the GUI produce for the bootstrapping?

As you draft your analysis is is good practice to exit the GUI via <paste> .


On 6/8/2012 12:57 PM, chutson wrote:

I have conducted an RCT looking at an intervention for dementia.  I have two
groups - (control and experimental) and two sets of measures (pre and post).
I did want to perform an ANCOVA on the pre/post difference scores for the
two groups but some of the data is not normal. I was therefore going to try
bootstrapping:

Analyze > Descriptive Statistics > Explore...
Explore main dialog
Select difference score as a dependent variable.
Select Statistics in the Display group.
Click Bootstrap.
Select Perform bootstrapping.
select Set seed for Mersenne Twister and type
592004 as the seed.
To obtain more accurate intervals (at the cost of more processing time),
select Bias corrected
accelerated (BCa)
Click Continue.
Click OK in the Explore dialog box.


But then bootstrapping doesnt come up in the output - instead it says: some
or all bootstrap sample results are issing, so no bootstrap estimation has
been performed for this table.

Can anyone help?
Thanks
Charlotte

--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Bootstrapping-in-SPSS-19-tp5713597.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

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

[Art Kendall]

The GUI (menus, etc) is a great shortcut to generating syntax.  However, the syntax is the set of  commands that are passed to the program.
It is very useful for doing the quality assurance review, communicating exactly what you told the computer to do, getting help, etc.
The menus have a <paste> button.

As Matthew Poes said using difference scores can be problematical.  That is most likely more important than slight departures from normality.

You want the residuals to be normal, etc.
Usually a researcher wants the means to be different.  That means that an overall histogram would have a high place for the control group i.e., a mean and a different high place, i.e., mean for the treated group.

take a look at the simplest situation of a 2-group ANOVA called the t-test. http://www.socialresearchmethods.net/kb/stat_t.php

Try using the graphics to visualize you data so you have a better handle on it.



On 6/8/2012 1:34 PM, charlotte hutson wrote:

I used kurtosis and skewness statistics and examined the histograms - the departure from normality is not drastic but I wanted to be comprehensive and use bootstrapping aswell
 
I didnt use the syntax screen - I just used the bootstrapping button on spss 19
 
why would you want it to be bimodal?
 
thanks
C
 

---

Date: Fri, 8 Jun 2012 13:29:49 -0400
From: [hidden email]
To: [hidden email]
CC: [hidden email]
Subject: Re: [SPSSX-L] Bootstrapping in SPSS 19

How did you decide something was not normal?
Imagine a graphic with difference score on the horizontal axis and Frequency on the vertical axis.
One would want (hope, hypothesize) that to be bi-modal. 
In these General Linear Models of which ANCOVA is a special instance, one worries whether the _residuals_ are flagrantly disparate from normal.

Clearly 2 variables are quite non-normal. The groups that cases were randomized to is described by a dichotomous variable treatment vs control and time is described by pre vs post.  However these are on the independent/predictor/right-hand side.

What syntax did you use to do the original analysis?

What syntax did your use of the GUI produce for the bootstrapping?

As you draft your analysis is is good practice to exit the GUI via <paste> .


On 6/8/2012 12:57 PM, chutson wrote:

I have conducted an RCT looking at an intervention for dementia.  I have two
groups - (control and experimental) and two sets of measures (pre and post).
I did want to perform an ANCOVA on the pre/post difference scores for the
two groups but some of the data is not normal. I was therefore going to try
bootstrapping:

Analyze > Descriptive Statistics > Explore...
Explore main dialog
Select difference score as a dependent variable.
Select Statistics in the Display group.
Click Bootstrap.
Select Perform bootstrapping.
select Set seed for Mersenne Twister and type
592004 as the seed.
To obtain more accurate intervals (at the cost of more processing time),
select Bias corrected
accelerated (BCa)
Click Continue.
Click OK in the Explore dialog box.


But then bootstrapping doesnt come up in the output - instead it says: some
or all bootstrap sample results are issing, so no bootstrap estimation has
been performed for this table.

Can anyone help?
Thanks
Charlotte

--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Bootstrapping-in-SPSS-19-tp5713597.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

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