R help - Aug 2008 - Survey Design / Rake questions

I'm trying to learn how to calibrate/postStratify/rake survey data in
preparation for a large survey effort we're about to embark upon.  As a
working example, I have results from a small survey of ~650 respondents,
~90 response fields each.  I'm trying to learn how to (properly?) apply
the aforementioned functions.  

 

 

My data are from a bus on board survey.  The expansion in the dataset is
derived from three elements:

  Response rates by bus stop for a sampled run

  Total runs/samples runs 

  Normalized to (separately derived) daily line boarding  

  

  

In order to get to the point of raking the data, I need to learn more
about the survey package and nomenclature.  For instance, given how I've
described the survey/weighting, is my call to svydesign correct?  I'm
not sure I understand just what a "survey design" is.  Where can I
read
up on this?  What's a good reference for such things as "PSUs",
"cluster
sampling", and so on.  I've tried the following code, which fails:

 

 
> SurveyData <- read.spss("C:/Data/R/orange_delivery.sav",
use.value.labels=TRUE, max.value.labels=Inf, to.data.frame=TRUE)
>
#=============================================================================
> temp <- sub(' +$', '', SurveyData$direction_) 
> SurveyData$direction_ <- temp
>
#=============================================================================
>
SurveyData$NumStn=abs(as.numeric(SurveyData$lineon)-as.numeric(SurveyDat
a$lineoff))
> EBSurvey <- subset(SurveyData, direction_ == "EASTBOUND" )
> XTTable <- xtabs(~direction_ , EBSurvey)
> XTTable
direction_

EASTBOUND 

      345 
> WBSurvey <- subset(SurveyData, direction_ == "WESTBOUND" )
> XTTable <- xtabs(~direction_ , WBSurvey)
> XTTable
direction_

WESTBOUND 

      307 
> #
> EBDesign <- svydesign(id=~sampn, weights=~expwgt, data=EBSurvey)
> #   svytable(~lineon+lineoff, EBDesign)
> OnLabels    <- c( "Warner Center", "De Soto",
"Pierce College",
"Tampa", "Reseda", "Balboa", "Woodley",
"Sepulveda", "Van Nuys",
"Woodman", "Valley College", "Laurel Canyon",
"North Hollywood")
> EBOnNewTots <- c(            1000,       600,             1200,
500,     1000,      500,       200,         250,       1000,       300,
100,              50,            73.65 )
> EBNumStn <- c(673.65,     800, 1000, 1000,  800,  700,  600, 500, 400,
200,  50, 50 )
> ByEBOn <- data.frame(OnLabels,EBOnNewTots)
> ByEBNum <- data.frame(c(1:12),EBNumStn)
> RakedEBSurvey <- rake(EBDesign, list(~ByEBOn, ~ByEBNum),
list(EBOnNewTots, EBNumStn ) )

Error in model.frame.default(margin, data = design$variables) : 

  invalid type (list) for variable 'ByEBOn'
> 
 

 

 

Robert Farley

Metro

1 Gateway Plaza

Mail Stop 99-23-7

Los Angeles, CA 90012-2952

Voice: (213)922-2532

Fax:    (213)922-2868

www.Metro.net 

 

 


	[[alternative HTML version deleted]]

Your reading, in increasing order of difficulty/mathematical details,
might be Lohr's "Sampling"
(http://www.citeulike.org/user/ctacmo/article/1068825), Korn &
Graubard's "Health Surveys"
(http://www.citeulike.org/user/ctacmo/article/553280), and Sarndal et.
al. Survey Math Bible
(http://www.citeulike.org/user/ctacmo/article/716032). You certainly
should try to get a hold of the primary concepts before collecting
your data (or rather before designing your survey... so it might
already be too late!). Post-stratification is not that huge topic, for
some reason; a review of mathematical details is given by Valliant
(1993) (http://www.citeulike.org/user/ctacmo/article/1036976). On
raking, the paper on top of Google Scholar search by Deville, Sarndal
and Sautory (1993)
(http://www.citeulike.org/user/ctacmo/article/3134001) is certainly
coming from the best people in the field.

I am not aware of general treatment of transportation survey sampling,
although I suspect such references do exist in transportation
research. There might be particular twists as the same subject/bus
usage episode might be sampled at different locations.

As far as rake() procedure is concerned, you need to have your data
set up as sampled observations with two classifications across which
you will be raking, probably the directions "E"/"W" and the
stations.
Those are not different data.frames, as you are trying to set them up,
but a single data.frame with several columns. In other words, your
sampled data will have labels "E"/"W" in one of the columns,
and
station names in another column, and (the names of) those columns will
be the imputs of rake().

On 8/18/08, Farley, Robert <FarleyR at metro.net>
wrote:
> I'm trying to learn how to calibrate/postStratify/rake survey data in
>  preparation for a large survey effort we're about to embark upon.  As
a
>  working example, I have results from a small survey of ~650 respondents,
>  ~90 response fields each.  I'm trying to learn how to (properly?)
apply
>  the aforementioned functions.
>
>  My data are from a bus on board survey.  The expansion in the dataset is
>  derived from three elements:
>
>   Response rates by bus stop for a sampled run
>
>   Total runs/samples runs
>
>   Normalized to (separately derived) daily line boarding
>
>  In order to get to the point of raking the data, I need to learn more
>  about the survey package and nomenclature.  For instance, given how
I've
>  described the survey/weighting, is my call to svydesign correct?  I'm
>  not sure I understand just what a "survey design" is.  Where can
I read
>  up on this?  What's a good reference for such things as
"PSUs", "cluster
>  sampling", and so on.
-- 
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.