Loop with string variable AND customizable "summary" output

Dear All,

I am using R for my research and I have two questions about it:

1) is it possible to create a loop using a string, instead of a numeric
vector? I have in mind a specific problem:

Suppose you have 2 countries: UK, and USA, one dependent (y) and one
independent variable (y) for each country (vale a dire: yUK, xUK, yUSA,
xUSA) and you want to run automatically the following regressions:

 

for (i in c("UK","USA"))

output{i}<-summary(lm(y{i} ~ x{i}))

 

In other words, at the end I would like to have two objects as output:
"outputUK" and "outputUSA", which contain respectively the results of
the first and second regression (yUK on xUK and yUSA on xUSA).

 

2) in STATA there is a very nice code ("outreg") to display nicely (and
as the user wants to) your regression results.

Is there anything similar in R / R contributed packages? More precisely,
I am thinking of something that is close in spirit to "summary" but it
is also customizable. For example, suppose you want different Signif.
codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 or a different
format display (i.e. without "t value" column) implemented automatically
(without manually editing it every time).

In alternative, if I was able to see it, I could modify the source code
of the function "summary", but I am not able to see its (line by line)
code. Any idea?

Or may be a customizable regression output already exists?

Thanks really a lot!

Carlo

Dear All,

I am using R for my research and I have two questions about it:

1) is it possible to create a loop using a string, instead of a numeric vector? I have in mind a specific problem:

Suppose you have 2 countries: UK, and USA, one dependent (y) and one independent variable (y) for each country (vale a dire: yUK, xUK, yUSA, xUSA) and you want to run automatically the following regressions:



for (i in c("UK","USA"))

output{i}<-summary(lm(y{i} ~ x{i}))



In other words, at the end I would like to have two objects as output: "outputUK" and "outputUSA", which contain respectively the results of the first and second regression (yUK on xUK and yUSA on xUSA).



2) in STATA there is a very nice code ("outreg") to display nicely (and as the user wants to) your regression results.

Is there anything similar in R / R contributed packages? More precisely, I am thinking of something that is close in spirit to "summary" but it is also customizable. For example, suppose you want different Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 or a different format display (i.e. without "t value" column) implemented automatically (without manually editing it every time).

In alternative, if I was able to see it, I could modify the source code of the function "summary", but I am not able to see its (line by line) code. Any idea?

Or may be a customizable regression output already exists?

Thanks really a lot!

Carlo

Dear All,

I am using R for my research and I have two questions about it:

1) is it possible to create a loop using a string, instead of a numeric
vector? I have in mind a specific problem:

for (i in c("UK","USA"))

output{i}<-summary(lm(y{i} ~ x{i}))

In other words, at the end I would like to have two objects as output:
"outputUK" and "outputUSA", which contain respectively the results of the
first and second regression (yUK on xUK and yUSA on xUSA). 

2)  I am thinking of something that is close in spirit to "summary" but it
is also customizable. For example, suppose you want different Signif.
codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 or a different
format display (i.e. without "t value" column) implemented automatically
(without manually editing it every time).

In alternative, if I was able to see it, I could modify the source code of
the function "summary", but I am not able to see its (line by line) code.
Any idea?

methods(summary)
methods(print)

summary(outputUK)

for (i in c("UK","USA"))
output{i}<-summary(lm(y{i} ~ x{i}))

That is

C.Rosa wrote:

for (i in c("UK","USA"))
output{i}<-summary(lm(y{i} ~ x{i}))

for (i in c("UK","USA")) {
  lm.txt<-paste("output",i,"<-","lm(","y",i,"~","x",i,")",sep="") # 1.
produce a character string containing needed expression
  eval(parse(text=lm.txt))                                                
# 2. parse and evaluate it
}

Dear All,

I am using R for my research and I have two questions about it:

1) is it possible to create a loop using a string, instead of a numeric vector? I have in mind a specific problem:

Suppose you have 2 countries: UK, and USA, one dependent (y) and one independent variable (y) for each country (vale a dire: yUK, xUK, yUSA, xUSA) and you want to run automatically the following regressions:



for (i in c("UK","USA"))

output{i}<-summary(lm(y{i} ~ x{i}))



In other words, at the end I would like to have two objects as output: "outputUK" and "outputUSA", which contain respectively the results of the first and second regression (yUK on xUK and yUSA on xUSA).



2) in STATA there is a very nice code ("outreg") to display nicely (and as the user wants to) your regression results.

Is there anything similar in R / R contributed packages? More precisely, I am thinking of something that is close in spirit to "summary" but it is also customizable. For example, suppose you want different Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 or a different format display (i.e. without "t value" column) implemented automatically (without manually editing it every time).

In alternative, if I was able to see it, I could modify the source code of the function "summary", but I am not able to see its (line by line) code. Any idea?

Or may be a customizable regression output already exists?

Thanks really a lot!

Carlo

Dear All,

I am using R for my research and I have two questions about it:

1) is it possible to create a loop using a string, instead of a numeric vector? I have in mind a specific problem:

Suppose you have 2 countries: UK, and USA, one dependent (y) and one independent variable (y) for each country (vale a dire: yUK, xUK, yUSA, xUSA) and you want to run automatically the following regressions:



for (i in c("UK","USA"))

output{i}<-summary(lm(y{i} ~ x{i}))



In other words, at the end I would like to have two objects as output: "outputUK" and "outputUSA", which contain respectively the results of the first and second regression (yUK on xUK and yUSA on xUSA).



2) in STATA there is a very nice code ("outreg") to display nicely (and as the user wants to) your regression results.

Is there anything similar in R / R contributed packages? More precisely, I am thinking of something that is close in spirit to "summary" but it is also customizable. For example, suppose you want different Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 or a different format display (i.e. without "t value" column) implemented automatically (without manually editing it every time).

In alternative, if I was able to see it, I could modify the source code of the function "summary", but I am not able to see its (line by line) code. Any idea?

Or may be a customizable regression output already exists?

Thanks really a lot!

Carlo

Often you will find that if you arrange your data in a
desirable way in the first place everything becomes
easier.  What you really want is a data frame such
as the last three columns of the builtin data frame
CO2 where Treatment corresponds to country and
the two numeric variables correspond to your y and x.

Then its easy:

lapply(levels(CO2$Treatment), function(lev)
  lm(uptake ~ conc, CO2, subset = Treatment == lev))

The only problem with the above is that the Call: in the
output does not really tell you which level of Treatment
is being used since it literally shows
 "lm(uptake ~ conc, CO2, subset = Treatment == lev)"
each time.  To get around substitute the value of lev in.
Because R uses delayed evaluation you also need to force the
evaluation of lev prior to substituting it in:

lapply(levels(CO2$Treatment), function(lev) {
  lev <- force(lev)
  eval(substitute(lm(uptake ~ conc, CO2, subset = Treatment == lev)),
    list(lev = lev))
})


Now if you really want to do it the way you specified originally
try this.

Suppose we use attach to grab the variables
x1, x2, x3, x4, y1, y2, y3, y4 out of the builtin
anscombe data frame for purposes of getting
our hands on some sample data.   In your case
the variables would already be in the workspace
so the attach is not needed.

Then simply reconstruct the formula in fo.  You
could simply use lm(fo) but then the Call: in the
output of lm would literally read lm(fo) so its
better to use do.call:

# next line gives the variables x1, x2, x3, x4, y1, y2, y3, y4
# from the builtin ancombe data set.
# In your case such variables would already exist.
attach(anscombe)
lapply(1:4, function(i) {
  ynm <- paste("y", i, sep = "")
  xnm <- paste("x", i, sep = "")
  fo <- as.formula(paste(ynm, "~", xnm))
  do.call("lm", list(fo))
})
detach(anscombe)

Or if all the variables have the same length you could use
a form such as ancombe in the first place:

Actually this is not really a recommended way of
proceeding. You would be better off putting all
your variables in a data frame and using that.

lapply(1:4, function(i) {
   fo <- as.formula(paste(names(anscombe)[i+4], "~", names(anscombe)[i]))
   do.call("lm", list(fo, data = quote(anscombe)))
})

or

lapply(1:4, function(i) {
   fo <- y ~ x
   fo[[2]] <- as.name(names(anscombe)[i+4])
   fo[[3]] <- as.name(names(anscombe)[i])
   do.call("lm", list(fo, data = quote(anscombe)))
})



On 1/29/07, C.Rosa at lse.ac.uk <C.Rosa at lse.ac.uk> wrote:

Dear All,

I am using R for my research and I have two questions about it:

1) is it possible to create a loop using a string, instead of a numeric vector? I have in mind a specific problem:

Suppose you have 2 countries: UK, and USA, one dependent (y) and one independent variable (y) for each country (vale a dire: yUK, xUK, yUSA, xUSA) and you want to run automatically the following regressions:



for (i in c("UK","USA"))

output{i}<-summary(lm(y{i} ~ x{i}))



In other words, at the end I would like to have two objects as output: "outputUK" and "outputUSA", which contain respectively the results of the first and second regression (yUK on xUK and yUSA on xUSA).



2) in STATA there is a very nice code ("outreg") to display nicely (and as the user wants to) your regression results.

Is there anything similar in R / R contributed packages? More precisely, I am thinking of something that is close in spirit to "summary" but it is also customizable. For example, suppose you want different Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 or a different format display (i.e. without "t value" column) implemented automatically (without manually editing it every time).

In alternative, if I was able to see it, I could modify the source code of the function "summary", but I am not able to see its (line by line) code. Any idea?

Or may be a customizable regression output already exists?

Thanks really a lot!

Carlo

Dear All,

I am using R for my research and I have two questions about it:

1) is it possible to create a loop using a string, instead of a numeric

Suppose you have 2 countries: UK, and USA, one dependent (y) and one

for (i in c("UK","USA"))

output{i}<-summary(lm(y{i} ~ x{i}))



In other words, at the end I would like to have two objects as output:

2) in STATA there is a very nice code ("outreg") to display nicely (and as

Is there anything similar in R / R contributed packages? More precisely, I

In alternative, if I was able to see it, I could modify the source code of

Or may be a customizable regression output already exists?

Thanks really a lot!

Carlo

In thinking about this a bit more here is an even shorter one yet it
does show the level in the Call output.  See ?bquote

lapply(levels(CO2$Treatment), function(lev)
  eval(bquote(lm(uptake ~ conc, CO2, subset = Treatment == .(lev)))))


On 1/29/07, Gabor Grothendieck <ggrothendieck at gmail.com> wrote:

Often you will find that if you arrange your data in a
desirable way in the first place everything becomes
easier.  What you really want is a data frame such
as the last three columns of the builtin data frame
CO2 where Treatment corresponds to country and
the two numeric variables correspond to your y and x.

Then its easy:

lapply(levels(CO2$Treatment), function(lev)
  lm(uptake ~ conc, CO2, subset = Treatment == lev))

The only problem with the above is that the Call: in the
output does not really tell you which level of Treatment
is being used since it literally shows
 "lm(uptake ~ conc, CO2, subset = Treatment == lev)"
each time.  To get around substitute the value of lev in.
Because R uses delayed evaluation you also need to force the
evaluation of lev prior to substituting it in:

lapply(levels(CO2$Treatment), function(lev) {
  lev <- force(lev)
  eval(substitute(lm(uptake ~ conc, CO2, subset = Treatment == lev)),
    list(lev = lev))
})


Now if you really want to do it the way you specified originally
try this.

Suppose we use attach to grab the variables
x1, x2, x3, x4, y1, y2, y3, y4 out of the builtin
anscombe data frame for purposes of getting
our hands on some sample data.   In your case
the variables would already be in the workspace
so the attach is not needed.

Then simply reconstruct the formula in fo.  You
could simply use lm(fo) but then the Call: in the
output of lm would literally read lm(fo) so its
better to use do.call:

# next line gives the variables x1, x2, x3, x4, y1, y2, y3, y4
# from the builtin ancombe data set.
# In your case such variables would already exist.
attach(anscombe)
lapply(1:4, function(i) {
  ynm <- paste("y", i, sep = "")
  xnm <- paste("x", i, sep = "")
  fo <- as.formula(paste(ynm, "~", xnm))
  do.call("lm", list(fo))
})
detach(anscombe)

Or if all the variables have the same length you could use
a form such as ancombe in the first place:

Actually this is not really a recommended way of
proceeding. You would be better off putting all
your variables in a data frame and using that.

lapply(1:4, function(i) {
   fo <- as.formula(paste(names(anscombe)[i+4], "~", names(anscombe)[i]))
   do.call("lm", list(fo, data = quote(anscombe)))
})

or

lapply(1:4, function(i) {
   fo <- y ~ x
   fo[[2]] <- as.name(names(anscombe)[i+4])
   fo[[3]] <- as.name(names(anscombe)[i])
   do.call("lm", list(fo, data = quote(anscombe)))
})



On 1/29/07, C.Rosa at lse.ac.uk <C.Rosa at lse.ac.uk> wrote:

Dear All,

I am using R for my research and I have two questions about it:

1) is it possible to create a loop using a string, instead of a numeric vector? I have in mind a specific problem:

Suppose you have 2 countries: UK, and USA, one dependent (y) and one independent variable (y) for each country (vale a dire: yUK, xUK, yUSA, xUSA) and you want to run automatically the following regressions:



for (i in c("UK","USA"))

output{i}<-summary(lm(y{i} ~ x{i}))



In other words, at the end I would like to have two objects as output: "outputUK" and "outputUSA", which contain respectively the results of the first and second regression (yUK on xUK and yUSA on xUSA).



2) in STATA there is a very nice code ("outreg") to display nicely (and as the user wants to) your regression results.

Is there anything similar in R / R contributed packages? More precisely, I am thinking of something that is close in spirit to "summary" but it is also customizable. For example, suppose you want different Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 or a different format display (i.e. without "t value" column) implemented automatically (without manually editing it every time).

In alternative, if I was able to see it, I could modify the source code of the function "summary", but I am not able to see its (line by line) code. Any idea?

Or may be a customizable regression output already exists?

Thanks really a lot!

Carlo

Prior answers are certainly correct, but this is where lists and lapply
shine:

result<-lapply(list(UK,USA),function(z)summary(lm(y~x,data=z)))

As in (nearly) all else, simplicity is a virtue.

If you prefer to keep the data sources as a character vector,dataNames,

result<-lapply(dataNames,function(z)summary(lm(y~x,data=get(z)))) 

should work. 

Note: both of these are untested for the general case where they might be
used within a function and may not find the right z unless you pay attention
to scope, especially in the get() construction.


Bert Gunter
Genentech Nonclinical Statistics
South San Francisco, CA 94404


-----Original Message-----
From: r-help-bounces at stat.math.ethz.ch
[mailto:r-help-bounces at stat.math.ethz.ch] On Behalf Of C.Rosa at lse.ac.uk
Sent: Monday, January 29, 2007 8:23 AM
To: liuwensui at gmail.com; bcarvalh at jhsph.edu; Roger.Bivand at nhh.no
Cc: r-help at stat.math.ethz.ch
Subject: Re: [R] Loop with string variable AND customizable "summary" output

Dear All,
Thank you very much for your help!
Carlo

-----Original Message-----
From: Wensui Liu [mailto:liuwensui at gmail.com]
Sent: Mon 29/01/2007 15:39
To: Rosa,C
Cc: r-help at stat.math.ethz.ch
Subject: Re: [R] Loop with string variable AND customizable "summary" output
 
Carlo,

try something like:

for (i in c("UK","USA"))
{
summ<-summary(lm(y ~ x), subset = (country = i))
assign(paste('output', i, sep = ''), summ);
}

(note: it is untested, sorry).

On 1/29/07, C.Rosa at lse.ac.uk <C.Rosa at lse.ac.uk> wrote:

Dear All,

I am using R for my research and I have two questions about it:

1) is it possible to create a loop using a string, instead of a numeric

vector? I have in mind a specific problem:

Suppose you have 2 countries: UK, and USA, one dependent (y) and one

independent variable (y) for each country (vale a dire: yUK, xUK, yUSA,
xUSA) and you want to run automatically the following regressions:

for (i in c("UK","USA"))

output{i}<-summary(lm(y{i} ~ x{i}))



In other words, at the end I would like to have two objects as output:

"outputUK" and "outputUSA", which contain respectively the results of the
first and second regression (yUK on xUK and yUSA on xUSA).

2) in STATA there is a very nice code ("outreg") to display nicely (and as

the user wants to) your regression results.

Is there anything similar in R / R contributed packages? More precisely, I

am thinking of something that is close in spirit to "summary" but it is also
customizable. For example, suppose you want different Signif. codes:  0
'***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 or a different format display
(i.e. without "t value" column) implemented automatically (without manually
editing it every time).

In alternative, if I was able to see it, I could modify the source code of

the function "summary", but I am not able to see its (line by line) code.
Any idea?

Or may be a customizable regression output already exists?

Thanks really a lot!

Carlo

summary(lmList(uptake ~ conc | Treatment, CO2))

Or, to throw yet another couple of possibilities into the mix:

lapply(split(YourDF, YourDF$country),
      function(x) summary(lm(y ~ x, data = x))

and:

library(nlme)
summary(lmList(y ~ x | country, YourDF))


See ?split and help(lmList, package = nlme)

HTH,

Marc Schwartz

On Mon, 2007-01-29 at 09:03 -0800, Bert Gunter wrote:

Prior answers are certainly correct, but this is where lists and lapply
shine:

result<-lapply(list(UK,USA),function(z)summary(lm(y~x,data=z)))

As in (nearly) all else, simplicity is a virtue.

If you prefer to keep the data sources as a character vector,dataNames,

result<-lapply(dataNames,function(z)summary(lm(y~x,data=get(z))))

should work.

Note: both of these are untested for the general case where they might be
used within a function and may not find the right z unless you pay attention
to scope, especially in the get() construction.


Bert Gunter
Genentech Nonclinical Statistics
South San Francisco, CA 94404


-----Original Message-----
From: r-help-bounces at stat.math.ethz.ch
[mailto:r-help-bounces at stat.math.ethz.ch] On Behalf Of C.Rosa at lse.ac.uk
Sent: Monday, January 29, 2007 8:23 AM
To: liuwensui at gmail.com; bcarvalh at jhsph.edu; Roger.Bivand at nhh.no
Cc: r-help at stat.math.ethz.ch
Subject: Re: [R] Loop with string variable AND customizable "summary" output

Dear All,
Thank you very much for your help!
Carlo

-----Original Message-----
From: Wensui Liu [mailto:liuwensui at gmail.com]
Sent: Mon 29/01/2007 15:39
To: Rosa,C
Cc: r-help at stat.math.ethz.ch
Subject: Re: [R] Loop with string variable AND customizable "summary" output

Carlo,

try something like:

for (i in c("UK","USA"))
{
summ<-summary(lm(y ~ x), subset = (country = i))
assign(paste('output', i, sep = ''), summ);
}

(note: it is untested, sorry).

On 1/29/07, C.Rosa at lse.ac.uk <C.Rosa at lse.ac.uk> wrote:

Dear All,

I am using R for my research and I have two questions about it:

1) is it possible to create a loop using a string, instead of a numeric

vector? I have in mind a specific problem:

Suppose you have 2 countries: UK, and USA, one dependent (y) and one

independent variable (y) for each country (vale a dire: yUK, xUK, yUSA,
xUSA) and you want to run automatically the following regressions:

for (i in c("UK","USA"))

output{i}<-summary(lm(y{i} ~ x{i}))



In other words, at the end I would like to have two objects as output:

"outputUK" and "outputUSA", which contain respectively the results of the
first and second regression (yUK on xUK and yUSA on xUSA).

2) in STATA there is a very nice code ("outreg") to display nicely (and as

the user wants to) your regression results.

Is there anything similar in R / R contributed packages? More precisely, I

am thinking of something that is close in spirit to "summary" but it is also
customizable. For example, suppose you want different Signif. codes:  0
'***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 or a different format display
(i.e. without "t value" column) implemented automatically (without manually
editing it every time).

In alternative, if I was able to see it, I could modify the source code of

the function "summary", but I am not able to see its (line by line) code.
Any idea?

Or may be a customizable regression output already exists?

Thanks really a lot!

Carlo

Note that the nlme solution seems to give the same coefficients
but appears to use a single error term rather than one error
term per level of the conditioning variable and that would change various
other statistics relative to the other solutions should that matter.

summary(lmList(uptake ~ conc | Treatment, CO2))

Call:
  Model: uptake ~ conc | Treatment
   Data: CO2

Coefficients:
   (Intercept)
           Estimate Std. Error  t value     Pr(>|t|)
nonchilled 22.01916    2.46416 8.935769 1.174616e-13
chilled    16.98142    2.46416 6.891361 1.146556e-09
   conc
             Estimate  Std. Error  t value     Pr(>|t|)
nonchilled 0.01982458 0.004692544 4.224699 6.292679e-05
chilled    0.01563659 0.004692544 3.332221 1.306259e-03

Residual standard error: 8.945667 on 80 degrees of freedom

summary(lmList(uptake ~ conc | Treatment, CO2, pool = FALSE))