Advice on recoding a variable depending on another which contains NAs

Dear colleagues,

I would be very grateful for your help with the following. I have  
banged my head off this question several times in the past, and  
repeatedly over the last week. I have looked in the usual places and  
found no obvious solution. I fear that this just means I didn't  
recognize it, but I'd be very grateful for your help.

I am scoring 8000 psychometric tests - the SCQ, if you have heard of  
it. On this test the scoring rules depends on one variable SCQ1 - if  
this is answered yes, the final score is a function of 39 variables,  
and if no, of 31 variables.

I've calculated both of these scores (SCQScore1 and SCQScore2)for  
all the children in my study, and I wish to create a final score,  
which is SCQScore1 when SCQ1 is 1, and SCQScore2 when SCQ1 is 2.  
There are also missing values for SCQ1, and I have chosen, for the  
moment, to set the final score to SCQScore1 for these. [[This is a  
debatable choice, but I am not asking your advice on that choice!]]

d$SCQScore <- 99
	##Distinct value for any other values I've missed

d$SCQScore[SCQ1 == 1] <- d$SCQScore1[SCQ1 == 1]
	## Talks using phrases/sentences, so sum S2CQ:SCQ40

d$SCQScore[SCQ1 == 2] <- d$SCQScore2[SCQ1 == 2]
	## Can't do this, so sum SCQ8:SCQ40

d$SCQScore[is.na(d$SCQ1)] <- d$SCQScore1 [is.na(d$SCQ1)]
	## SCQ1 is missing

This fails on line 2
(d$SCQScore[SCQ1 == 1] <- d$SCQScore1[SCQ1 == 1])
with the error message
"NAs are not allowed in subscripted assignments",
presumably because SCQ1 does indeed contain missing values.

This can be fixed, got around, or otherwise bypassed, by creating a  
new variable SCQ1, with no missing values, as shown :-

SCQ1 <- d$SCQ1
SCQ1[is.na(SCQ1)] <- 3

d$SCQScore[SCQ1 == 1] <- d$SCQScore1[SCQ1 == 1]
	## Talks using phrases/sentences so sum S2CQ:SCQ40
d$SCQScore[SCQ1 == 2] <- d$SCQScore2[SCQ1 == 2]
	## Can't do this, so sum SCQ8:SCQ40
d$SCQScore[SCQ1 == 3] <- d$SCQScore1[SCQ1 == 3]
	## We don't know if he/she can talk, so guess - sum S2:S40

This type of thing is a common problem in my little world. Is there  
a better/less klutzy/smarter way of solving it than creating a new  
variable each time? Please bear in mind that it is critical, for  
later analysis, to keep the missing values in SCQ1.

Best wishes,
Anthony Staines
-- 
Anthony Staines, Professor of Health Systems,
School of Nursing and Human Sciences, DCU, Dublin 9,Ireland.
Tel:- +353 1 700 7807. Mobile:- +353 86 606 9713
http://astaines.eu/

On Nov 19, 2011, at 6:31 PM, Anthony Staines wrote:


This would seem to be an obvious task for ifelse()

SCQScore <- NA
d$SCQScore <- ifelse( SCQ1 == 1, d$SCQScore1, d$SCOScore2)

(And don't use 99 for missing. Use NA. It will protect you
better than "99".)


I suppose you could enforce the two level testing with:

d$SCQScore <- ifelse( SCQ1 == 1, d$SCQScore1,
ifelse(SCQ1 ==2, d$SCOScore2, NA))

d$SCQScore <- 99
##Distinct value for any other values I've missed

d$SCQScore[SCQ1 == 1] <- d$SCQScore1[SCQ1 == 1]
## Talks using phrases/sentences, so sum S2CQ:SCQ40

d$SCQScore[SCQ1 == 2] <- d$SCQScore2[SCQ1 == 2]
## Can't do this, so sum SCQ8:SCQ40

d$SCQScore[is.na(d$SCQ1)] <- d$SCQScore1 [is.na(d$SCQ1)]
## SCQ1 is missing

This fails on line 2
(d$SCQScore[SCQ1 == 1] <- d$SCQScore1[SCQ1 == 1])
with the error message
"NAs are not allowed in subscripted assignments",
presumably because SCQ1 does indeed contain missing values.

This can be fixed, got around, or otherwise bypassed, by
creating a new variable SCQ1, with no missing values, as
shown :-

SCQ1 <- d$SCQ1
SCQ1[is.na(SCQ1)] <- 3

d$SCQScore[SCQ1 == 1] <- d$SCQScore1[SCQ1 == 1]
## Talks using phrases/sentences so sum S2CQ:SCQ40
d$SCQScore[SCQ1 == 2] <- d$SCQScore2[SCQ1 == 2]
## Can't do this, so sum SCQ8:SCQ40
d$SCQScore[SCQ1 == 3] <- d$SCQScore1[SCQ1 == 3]
## We don't know if he/she can talk, so guess - sum S2:S40

This type of thing is a common problem in my little world.
Is there a better/less klutzy/smarter way of solving it
than creating a new variable each time? Please bear in
mind that it is critical, for later analysis, to keep the
missing values in SCQ1.

Best wishes,
Anthony Staines
--
Anthony Staines, Professor of Health Systems,
School of Nursing and Human Sciences, DCU, Dublin 9,Ireland.
Tel:- +353 1 700 7807. Mobile:- +353 86 606 9713
http://astaines.eu/

David Winsemius, MD
West Hartford, CT

Ah,
so that's how ifelse gets used...

Presumably if I had more than 2 non-missing values in the 
control variable, I could use it several times.

Thank you very much, for a really useful answer, and thanks 
for getting back so quickly!

All the best,
Anthony Staines

On 11/19/11 23:55, David Winsemius wrote:

On Nov 19, 2011, at 6:31 PM, Anthony Staines wrote:


This would seem to be an obvious task for ifelse()

SCQScore <- NA
d$SCQScore <- ifelse( SCQ1 == 1, d$SCQScore1, d$SCOScore2)

(And don't use 99 for missing. Use NA. It will protect you
better than "99".)


I suppose you could enforce the two level testing with:

d$SCQScore <- ifelse( SCQ1 == 1, d$SCQScore1,
ifelse(SCQ1 ==2, d$SCOScore2, NA))

d$SCQScore <- 99
##Distinct value for any other values I've missed

d$SCQScore[SCQ1 == 1] <- d$SCQScore1[SCQ1 == 1]
## Talks using phrases/sentences, so sum S2CQ:SCQ40

d$SCQScore[SCQ1 == 2] <- d$SCQScore2[SCQ1 == 2]
## Can't do this, so sum SCQ8:SCQ40

d$SCQScore[is.na(d$SCQ1)] <- d$SCQScore1 [is.na(d$SCQ1)]
## SCQ1 is missing

This fails on line 2
(d$SCQScore[SCQ1 == 1] <- d$SCQScore1[SCQ1 == 1])
with the error message
"NAs are not allowed in subscripted assignments",
presumably because SCQ1 does indeed contain missing values.

This can be fixed, got around, or otherwise bypassed, by
creating a new variable SCQ1, with no missing values, as
shown :-

SCQ1 <- d$SCQ1
SCQ1[is.na(SCQ1)] <- 3

d$SCQScore[SCQ1 == 1] <- d$SCQScore1[SCQ1 == 1]
## Talks using phrases/sentences so sum S2CQ:SCQ40
d$SCQScore[SCQ1 == 2] <- d$SCQScore2[SCQ1 == 2]
## Can't do this, so sum SCQ8:SCQ40
d$SCQScore[SCQ1 == 3] <- d$SCQScore1[SCQ1 == 3]
## We don't know if he/she can talk, so guess - sum S2:S40

This type of thing is a common problem in my little world.
Is there a better/less klutzy/smarter way of solving it
than creating a new variable each time? Please bear in
mind that it is critical, for later analysis, to keep the
missing values in SCQ1.

Best wishes,
Anthony Staines
--
Anthony Staines, Professor of Health Systems,
School of Nursing and Human Sciences, DCU, Dublin 9,Ireland.
Tel:- +353 1 700 7807. Mobile:- +353 86 606 9713
http://astaines.eu/

David Winsemius, MD
West Hartford, CT

-- 
Anthony Staines, Professor of Health Systems,
School of Nursing and Human Sciences, DCU, Dublin 9,Ireland.
Tel:- +353 1 700 7807. Mobile:- +353 86 606 9713
http://astaines.eu/