Bootstrap error message

Hi,

I'm trying to get a bootstrap estimate for confidence intervals of my regression parameters, but I keep getting the same error message "some or all of the bootstrap samples were missing, so no bootstrap sample was calculated". I've ensured there is no missing data, and still no luck...Is there anything else I can do?

[SPSS Support-2]

Hi,

Can you give more information? Is this a linear regression, with complete data (no missing values on any of the variables named on the command/specied in the dialog box)? How many cases are there? How many predictor (independent) variables? What release are you using? Are you getting the basic regression results?

David Nichols
IBM SPSS Support

From:	"[hidden email]" <[hidden email]>
To:	[hidden email]
Date:	10/26/2010 02:51 AM
Subject:	Bootstrap error message
Sent by:	"SPSSX(r) Discussion" <[hidden email]>

---

Hi,

I'm trying to get a bootstrap estimate for confidence intervals of my
regression parameters, but I keep getting the same error message "some or
all of the bootstrap samples were missing, so no bootstrap sample was
calculated". I've ensured there is no missing data, and still no luck...Is
there anything else I can do?
--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Bootstrap-error-message-tp3236664p3236664.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

[Nirit Avnimelech]

Hi,

I'm trying to produce tables that display - side by side - 2 or more
variables that have the same labels.
This is the syntax I'm using:


CTABLES
  /VLABELS VARIABLES= v1 v2 DISPLAY=label
  /TABLE BY v1 [C][COLPCT.COUNT '' PCT40.1, TOTALS[COLPCT.COUNT '' PCT40.1,
UCOUNT '' F40.0]]
    + v2 [C][COLPCT.COUNT '' PCT40.1, TOTALS[COLPCT.COUNT '' PCT40.1, UCOUNT
'' F40.0]]
  /SLABELS POSITION=ROW
  /CLABELS COLLABELS=OPPOSITE
  /CATEGORIES VARIABLES=  v1 v2 ORDER=a KEY= value EMPTY=INCLUDE TOTAL=YES
LABEL='TOTAL' POSITION=AFTER

I need to sort values by descending counts of the first variable (v1).
However, if I use the subcommand:

ORDER=d KEY= count(v1)

I get an error message claiming that in this type of table sorting is not
possible.
Is there any workaround?  Any script that can sort the existing table?

Thanks!
Nirit

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

[Jon Fry-2]

It's likely that in some of your bootstrap samples, one or more predictor variables became degenerate.  It had the same value in all the sampled cases, or in the sampled cases, it was a linear combination of other predictor variables.

One solution in linear regression problems is to sample residuals instead of cases.  To do that, run your regression first saving either residuals or predicted values.  Then run the BOOTSTRAP command as modified below, and then the same REGRESSION command.

To sample residuals from the full set of residuals, add

      /SAMPLING METHOD=RESIDUAL(RESIDUALS=<residual variable name>)

to the BOOTSTRAP command.  If you're not comfortable with the assumption that the residuals of all the cases come from a common distribution, use METHOD=WILD instead.  That reverses the direction of the residuals in about half the cases chosen at random.  The GUI does not offer these options.

Jonathan Fry


On 10/26/2010 2:49 AM, [hidden email] wrote:

Hi,

I'm trying to get a bootstrap estimate for confidence intervals of my
regression parameters, but I keep getting the same error message "some or
all of the bootstrap samples were missing, so no bootstrap sample was
calculated". I've ensured there is no missing data, and still no luck...Is
there anything else I can do?
--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Bootstrap-error-message-tp3236664p3236664.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

[raesun]

I am using a bootstrap procedure with a multiple regression for a sample of 90. I standardized the values before entering into the regression because I had so many mixed measures. I didn't have that much non-normality in my dataset, but I am getting this error message also after trying 5,000 and 10,000 samples. I am trying the residuals method using the standardized residuals. I have been waiting for it to run for 20 minutes.  Should I be using the standardized residuals in this case or something else?

Help asap would be appreciated - I am supposed to get a draft to my committee tonight.

I used the percentile method and got results, but get the error message with the BCA. I am leaning towards going the percentile route so that I can just get something out but wanted to ask for help.

Thanks,

rae

[raesun]

I got the error message using the BCA with 5000 and 10000 and the resampling syntax that the previous poster recommended. I didn't get it with the percentile method. What do I do?
 

raesun wrote

I am using a bootstrap procedure with a multiple regression for a sample of 90. I standardized the values before entering into the regression because I had so many mixed measures. I didn't have that much non-normality in my dataset, but I am getting this error message also after trying 5,000 and 10,000 samples. I am trying the residuals method using the standardized residuals. I have been waiting for it to run for 20 minutes.  Should I be using the standardized residuals in this case or something else?

[Alex Reutter]

What is the text of the error message you're getting, and can you post the syntax you're running?

Alex

From:	raesun <[hidden email]>
To:	[hidden email]
Date:	01/18/2011 08:41 AM
Subject:	Re: Bootstrap error message
Sent by:	"SPSSX(r) Discussion" <[hidden email]>

---

I am using a bootstrap procedure with a multiple regression for a sample of
90. I standardized the values before entering into the regression because I
had so many mixed measures. I didn't have that much non-normality in my
dataset, but I am getting this error message also after trying 5,000 and
10,000 samples. I am trying the residuals method using the standardized
residuals. I have been waiting for it to run for 20 minutes.  Should I be
using the standardized residuals in this case or something else?

Help asap would be appreciated - I am supposed to get a draft to my
committee tonight.

I used the percentile method and got results, but get the error message with
the BCA. I am leaning towards going the percentile route so that I can just
get something out but wanted to ask for help.

Thanks,

rae