bootstrapping quantile regression

3 messages · Roger Koenker, Frank E Harrell Jr, David Freedman

Original

1

3

Wed, Oct 31, 2012 5:52 AM #

There is no automatic "clustering" option for QR bootstrapping.
You will have to roll your own.


url:    www.econ.uiuc.edu/~roger            Roger Koenker
email    rkoenker at uiuc.edu            Department of Economics
vox:     217-333-4558                University of Illinois
fax:       217-244-6678                Urbana, IL 61801

On Oct 31, 2012, at 1:38 AM, Kay Cichini wrote:

sry, I forgot to replace rlm() - but actually I tried both and the question
applies to both approaches..

Am 31.10.2012 00:19 schrieb "Kay Cichini" <kay.cichini at gmail.com>:

HI everyone,

I try to get some bootstrap CIs for coefficients obtained by quantile

regression. I have influencial values and thus switched to quantreg..

The data is clustered and within clusters the variance of my DV = 0..

Is this sensible for the below data? And what about the warnings?

Thanks in advance for any guidance,
Kay

dput(d)

structure(list(Porenfl?che = c(4990L, 7002L, 7558L, 7352L, 7943L,
7979L, 9333L, 8209L, 8393L, 6425L, 9364L, 8624L, 10651L, 8868L,
9417L, 8874L, 10962L, 10743L, 11878L, 9867L, 7838L, 11876L, 12212L,
8233L, 6360L, 4193L, 7416L, 5246L, 6509L, 4895L, 6775L, 7894L,
5980L, 5318L, 7392L, 7894L, 3469L, 1468L, 3524L, 5267L, 5048L,
1016L, 5605L, 8793L, 3475L, 1651L, 5514L, 9718L), P.Perimeter = c(2791.9,
3892.6, 3930.66, 3869.32, 3948.54, 4010.15, 4345.75, 4344.75,
3682.04, 3098.65, 4480.05, 3986.24, 4036.54, 3518.04, 3999.37,
3629.07, 4608.66, 4787.62, 4864.22, 4479.41, 3428.74, 4353.14,
4697.65, 3518.44, 1977.39, 1379.35, 1916.24, 1585.42, 1851.21,
1239.66, 1728.14, 1461.06, 1426.76, 990.388, 1350.76, 1461.06,
1376.7, 476.322, 1189.46, 1644.96, 941.543, 308.642, 1145.69,
2280.49, 1174.11, 597.808, 1455.88, 1485.58), P.Form = c(0.0903296,
0.148622, 0.183312, 0.117063, 0.122417, 0.167045, 0.189651, 0.164127,
0.203654, 0.162394, 0.150944, 0.148141, 0.228595, 0.231623, 0.172567,
0.153481, 0.204314, 0.262727, 0.200071, 0.14481, 0.113852, 0.291029,
0.240077, 0.161865, 0.280887, 0.179455, 0.191802, 0.133083, 0.225214,
0.341273, 0.311646, 0.276016, 0.197653, 0.326635, 0.154192, 0.276016,
0.176969, 0.438712, 0.163586, 0.253832, 0.328641, 0.230081, 0.464125,
0.420477, 0.200744, 0.262651, 0.182453, 0.200447), Durchl?ssigkeit =

c(6.3,

6.3, 6.3, 6.3, 17.1, 17.1, 17.1, 17.1, 119, 119, 119, 119, 82.4,
82.4, 82.4, 82.4, 58.6, 58.6, 58.6, 58.6, 142, 142, 142, 142,
740, 740, 740, 740, 890, 890, 890, 890, 950, 950, 950, 950, 100,
100, 100, 100, 1300, 1300, 1300, 1300, 580, 580, 580, 580), Gebiete =

structure(c(1L,

1L, 1L, 1L, 2L, 2L, 2L, 2L, 6L, 6L, 6L, 6L, 4L, 4L, 4L, 4L, 3L,
3L, 3L, 3L, 7L, 7L, 7L, 7L, 9L, 9L, 9L, 9L, 10L, 10L, 10L, 10L,
11L, 11L, 11L, 11L, 5L, 5L, 5L, 5L, 12L, 12L, 12L, 12L, 8L, 8L,
8L, 8L), .Label = c("6.3", "17.1", "58.6", "82.4", "100", "119",
"142", "580", "740", "890", "950", "1300"), class = "factor")), .Names =

c("Porenfl?che",

"P.Perimeter", "P.Form", "Durchl?ssigkeit", "Gebiete"), row.names =

c("1",

"2", "3", "4", "5", "6", "7", "8", "9", "10", "11", "12", "13",
"14", "15", "16", "17", "18", "19", "20", "21", "22", "23", "24",
"25", "26", "27", "28", "29", "30", "31", "32", "33", "34", "35",
"36", "37", "38", "39", "40", "41", "42", "43", "44", "45", "46",
"47", "48"), class = "data.frame")

## do quantile regression and bootstrap the coefficients, allowing for

clustered data

## by putting "Gebiet" as strata argument (?),
## dv variation within clusters/Gebiet = 0!
bs <- function(formula, data, indices) {
 d <- data[indices, ] # allows boot to select sample
 fit <- rlm(formula, data = d)
 return(coef(fit))
}

results <- boot(data = d, statistic = bs, strata = d$Gebiete,
               R = 199, formula = Durchl?ssigkeit ~ P.Perimeter + P.Form)

# get 99% confidence intervals
boot.ci(results, type="bca", index=1, conf = .99) # intercept
boot.ci(results, type="bca", index=2, conf = .99) # P.Perimeter
boot.ci(results, type="bca", index=3, conf = .99) # P.Form

--

Kay Cichini, MSc Biol

Grubenweg 22, 6071 Aldrans

E-Mail: kay.cichini at gmail.com
--

	[[alternative HTML version deleted]]

______________________________________________
R-help at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.

Frank E Harrell Jr

Wed, Oct 31, 2012 6:18 AM #

A piece of this is solved by the rms package's Rq and bootcov functions. 
-Frank

Roger Koenker-3 wrote

There is no automatic "clustering" option for QR bootstrapping.
You will have to roll your own.


url:    www.econ.uiuc.edu/~roger            Roger Koenker
email

rkoenker@

            Department of Economics
vox:     217-333-4558                University of Illinois
fax:       217-244-6678                Urbana, IL 61801

On Oct 31, 2012, at 1:38 AM, Kay Cichini wrote:

sry, I forgot to replace rlm() - but actually I tried both and the
question
applies to both approaches..

Am 31.10.2012 00:19 schrieb "Kay Cichini" &lt;

kay.cichini@

&gt;:

HI everyone,

I try to get some bootstrap CIs for coefficients obtained by quantile

regression. I have influencial values and thus switched to quantreg..

The data is clustered and within clusters the variance of my DV = 0..

Is this sensible for the below data? And what about the warnings?

Thanks in advance for any guidance,
Kay

dput(d)

structure(list(Porenfl?che = c(4990L, 7002L, 7558L, 7352L, 7943L,
7979L, 9333L, 8209L, 8393L, 6425L, 9364L, 8624L, 10651L, 8868L,
9417L, 8874L, 10962L, 10743L, 11878L, 9867L, 7838L, 11876L, 12212L,
8233L, 6360L, 4193L, 7416L, 5246L, 6509L, 4895L, 6775L, 7894L,
5980L, 5318L, 7392L, 7894L, 3469L, 1468L, 3524L, 5267L, 5048L,
1016L, 5605L, 8793L, 3475L, 1651L, 5514L, 9718L), P.Perimeter =
c(2791.9,
3892.6, 3930.66, 3869.32, 3948.54, 4010.15, 4345.75, 4344.75,
3682.04, 3098.65, 4480.05, 3986.24, 4036.54, 3518.04, 3999.37,
3629.07, 4608.66, 4787.62, 4864.22, 4479.41, 3428.74, 4353.14,
4697.65, 3518.44, 1977.39, 1379.35, 1916.24, 1585.42, 1851.21,
1239.66, 1728.14, 1461.06, 1426.76, 990.388, 1350.76, 1461.06,
1376.7, 476.322, 1189.46, 1644.96, 941.543, 308.642, 1145.69,
2280.49, 1174.11, 597.808, 1455.88, 1485.58), P.Form = c(0.0903296,
0.148622, 0.183312, 0.117063, 0.122417, 0.167045, 0.189651, 0.164127,
0.203654, 0.162394, 0.150944, 0.148141, 0.228595, 0.231623, 0.172567,
0.153481, 0.204314, 0.262727, 0.200071, 0.14481, 0.113852, 0.291029,
0.240077, 0.161865, 0.280887, 0.179455, 0.191802, 0.133083, 0.225214,
0.341273, 0.311646, 0.276016, 0.197653, 0.326635, 0.154192, 0.276016,
0.176969, 0.438712, 0.163586, 0.253832, 0.328641, 0.230081, 0.464125,
0.420477, 0.200744, 0.262651, 0.182453, 0.200447), Durchl?ssigkeit =

c(6.3,

6.3, 6.3, 6.3, 17.1, 17.1, 17.1, 17.1, 119, 119, 119, 119, 82.4,
82.4, 82.4, 82.4, 58.6, 58.6, 58.6, 58.6, 142, 142, 142, 142,
740, 740, 740, 740, 890, 890, 890, 890, 950, 950, 950, 950, 100,
100, 100, 100, 1300, 1300, 1300, 1300, 580, 580, 580, 580), Gebiete =

structure(c(1L,

1L, 1L, 1L, 2L, 2L, 2L, 2L, 6L, 6L, 6L, 6L, 4L, 4L, 4L, 4L, 3L,
3L, 3L, 3L, 7L, 7L, 7L, 7L, 9L, 9L, 9L, 9L, 10L, 10L, 10L, 10L,
11L, 11L, 11L, 11L, 5L, 5L, 5L, 5L, 12L, 12L, 12L, 12L, 8L, 8L,
8L, 8L), .Label = c("6.3", "17.1", "58.6", "82.4", "100", "119",
"142", "580", "740", "890", "950", "1300"), class = "factor")), .Names =

c("Porenfl?che",

"P.Perimeter", "P.Form", "Durchl?ssigkeit", "Gebiete"), row.names =

c("1",

"2", "3", "4", "5", "6", "7", "8", "9", "10", "11", "12", "13",
"14", "15", "16", "17", "18", "19", "20", "21", "22", "23", "24",
"25", "26", "27", "28", "29", "30", "31", "32", "33", "34", "35",
"36", "37", "38", "39", "40", "41", "42", "43", "44", "45", "46",
"47", "48"), class = "data.frame")

## do quantile regression and bootstrap the coefficients, allowing for

clustered data

## by putting "Gebiet" as strata argument (?),
## dv variation within clusters/Gebiet = 0!
bs <- function(formula, data, indices) {
 d <- data[indices, ] # allows boot to select sample
 fit <- rlm(formula, data = d)
 return(coef(fit))
}

results <- boot(data = d, statistic = bs, strata = d$Gebiete,
               R = 199, formula = Durchl?ssigkeit ~ P.Perimeter +
P.Form)

# get 99% confidence intervals
boot.ci(results, type="bca", index=1, conf = .99) # intercept
boot.ci(results, type="bca", index=2, conf = .99) # P.Perimeter
boot.ci(results, type="bca", index=3, conf = .99) # P.Form

--

Kay Cichini, MSc Biol

Grubenweg 22, 6071 Aldrans

E-Mail:

kay.cichini@

--

	[[alternative HTML version deleted]]

______________________________________________

https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.

______________________________________________

https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.

-----
Frank Harrell
Department of Biostatistics, Vanderbilt University
--
View this message in context: http://r.789695.n4.nabble.com/bootstrapping-quantile-regression-tp4647948p4648001.html
Sent from the R help mailing list archive at Nabble.com.

Wed, Oct 31, 2012 9:39 AM #

A possiblie solution might be to use the survey package.  You could specify
that the data is clustered using the svydesign function, and then speciy the
replicate weights using the as.svrepdesign function.  And then, it would be
possible to use the withReplicates function to bootstrap the clusters

A copy of Thomas Lumley's book - Complex Surveys - would probably help with
this

Hope this helps



--
View this message in context: http://r.789695.n4.nabble.com/bootstrapping-quantile-regression-tp4647948p4648032.html
Sent from the R help mailing list archive at Nabble.com.