27 Jul
2010
27 Jul
'10
12:14 p.m.
I have a question for the MI experts about imputations and experiments:
We often run experiments in which we hypothesize that responses will vary across
conditions. We expose subjects to a condition -- CONDa, CONDb, or CONDc, say -- and then
measure responses to a dependent variables across all conditions, say DV1, DV2, etc. We
also collect data on various independent variables, say IV1, IV2, etc.
But because we anticipate the relationship between the IVs and DVs to vary across the
conditions it seems like we ought to do one of two things when imputing missing data:
(1) interact every IV with the conditions so that we have, in effect DV1a, DV1_CONDb,
DV1_CONDc, DV2_CONDa, etc. But quite often, the result of that equation will just result
in a zero when the dummy for that condition is zero rather than one. This seems wasteful
of information to me. Which leads me to my alternative...
(2) Rather than computing DV_CONDa as equalling zero when CONDa is also zero, I'm
tempted to treat every alternative DV (DV1_CONDa, DV1_CONDb, DV1_CONDc ...) as missing for
each case and impute it. Missingness will be high, of course (only 1/conditions of DVs
will be present), but at least I won't be throwing away lots of valuable data. I
hesitate to do so because I can't find anyone else who has done this and that makes
me think I am probably misguided.
ps. I realize that this is a question about imputation generally, but I thought I'd
post it here since I use Amelia for my imputation needs -- let me know if I should not
post something like this here & I'll look elsewhere
Donald Braman
phone: 413-628-1221
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia
More info about Amelia: http://gking.harvard.edu/amelia
27 Jul
27 Jul
2:55 p.m.
both of these would 'work', and both have been done before lots of times I
think. The only issue is that Amelia isn't 'aware' of the structural
zeros
in the second case or the interactions in the first. e.g., for the
interactions, if you include A, B, and A*B as variables, it doesn't know
that the product of the first 2 variables is the 3rd and so the imputations
won't necessarily respect that known equality. The way we usually deal with
that is to do the imputation unconstrained with all 3 variables and then fix
it after the fact, such as by discarding A*B and multiplying the first two
columns together. This will usually give pretty reasonable results,
although there may be a new method someone could devise (and program!) that
could improve on this approach.
Gary
---
http://gking.harvard.edu
On Tue, Jul 27, 2010 at 12:14 PM, Donald Braman <donald.braman(a)gmail.com>wrote;wrote:
I have a question for the MI experts about imputations
and experiments:
We often run experiments in which we hypothesize that responses will vary
across conditions. We expose subjects to a condition -- CONDa, CONDb, or
CONDc, say -- and then measure responses to a dependent variables across all
conditions, say DV1, DV2, etc. We also collect data on various independent
variables, say IV1, IV2, etc.
But because we anticipate the relationship between the IVs and DVs to vary
across the conditions it seems like we ought to do one of two things when
imputing missing data:
(1) interact every IV with the conditions so that we have, in effect DV1a,
DV1_CONDb, DV1_CONDc, DV2_CONDa, etc. But quite often, the result of that
equation will just result in a zero when the dummy for that condition is
zero rather than one. This seems wasteful of information to me. Which
leads me to my alternative...
(2) Rather than computing DV_CONDa as equalling zero when CONDa is also
zero, I'm tempted to treat every alternative DV (DV1_CONDa, DV1_CONDb,
DV1_CONDc ...) as missing for each case and impute it. Missingness will be
high, of course (only 1/conditions of DVs will be present), but at least I
won't be throwing away lots of valuable data. I hesitate to do so because
I can't find anyone else who has done this and that makes me think I am
probably misguided.
ps. I realize that this is a question about imputation generally, but I
thought I'd post it here since I use Amelia for my imputation needs -- let
me know if I should not post something like this here & I'll look elsewhere
Donald Braman
phone: 413-628-1221
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia
More info about Amelia: http://gking.harvard.edu/amelia
28 Jul
28 Jul
6:32 p.m.
Thanks! Just to make sure: I'm talking about imputing all the dependent
variables for conditions to which that subject was *not* assigned, so if
there are 2, 3 or 4 conditions in all, respectively, 50%, 66% or 75% of the
data will be missing and imputed. It would almost be as if a
between-subjects experiment were turned into a within-subjects experiment as
we'd then have responses for every subject for every condition. Does that
make sense? -Don
On Tue, Jul 27, 2010 at 2:55 PM, Gary King <king(a)harvard.edu> wrote:
both of these would 'work', and both have
been done before lots of times I
think. The only issue is that Amelia isn't 'aware' of the structural
zeros
in the second case or the interactions in the first. e.g., for the
interactions, if you include A, B, and A*B as variables, it doesn't know
that the product of the first 2 variables is the 3rd and so the imputations
won't necessarily respect that known equality. The way we usually deal with
that is to do the imputation unconstrained with all 3 variables and then fix
it after the fact, such as by discarding A*B and multiplying the first two
columns together. This will usually give pretty reasonable results,
although there may be a new method someone could devise (and program!) that
could improve on this approach.
Gary
---
http://gking.harvard.edu
On Tue, Jul 27, 2010 at 12:14 PM, Donald Braman <donald.braman(a)gmail.com>wrote;wrote:
--
Donald Braman
phone: 971-645-0607
http://www.culturalcognition.net/braman/
http://ssrn.com/author=286206
http://www.law.gwu.edu/Faculty/profile.aspx?id=10123
I have a question for the MI experts about
imputations and experiments:
We often run experiments in which we hypothesize that responses will vary
across conditions. We expose subjects to a condition -- CONDa, CONDb, or
CONDc, say -- and then measure responses to a dependent variables across all
conditions, say DV1, DV2, etc. We also collect data on various independent
variables, say IV1, IV2, etc.
But because we anticipate the relationship between the IVs and DVs to vary
across the conditions it seems like we ought to do one of two things when
imputing missing data:
(1) interact every IV with the conditions so that we have, in effect DV1a,
DV1_CONDb, DV1_CONDc, DV2_CONDa, etc. But quite often, the result of that
equation will just result in a zero when the dummy for that condition is
zero rather than one. This seems wasteful of information to me. Which
leads me to my alternative...
(2) Rather than computing DV_CONDa as equalling zero when CONDa is also
zero, I'm tempted to treat every alternative DV (DV1_CONDa, DV1_CONDb,
DV1_CONDc ...) as missing for each case and impute it. Missingness will be
high, of course (only 1/conditions of DVs will be present), but at least I
won't be throwing away lots of valuable data. I hesitate to do so because
I can't find anyone else who has done this and that makes me think I am
probably misguided.
ps. I realize that this is a question about imputation generally, but I
thought I'd post it here since I use Amelia for my imputation needs -- let
me know if I should not post something like this here & I'll look elsewhere
Donald Braman
phone: 413-628-1221
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia
More info about Amelia: http://gking.harvard.edu/amelia
9:28 p.m.
yes, that makes sense. The results would normally be more model-dependent
because you're imputing a lot, but if you were running an experiment and
apply the conditions randomly the assumptions of the model are known to be
correct.
Gary
---
http://gking.harvard.edu
On Wed, Jul 28, 2010 at 6:32 PM, Donald Braman <dbraman(a)law.gwu.edu> wrote:
Thanks! Just to make sure: I'm talking about
imputing all the dependent
variables for conditions to which that subject was *not* assigned, so if
there are 2, 3 or 4 conditions in all, respectively, 50%, 66% or 75% of the
data will be missing and imputed. It would almost be as if a
between-subjects experiment were turned into a within-subjects experiment as
we'd then have responses for every subject for every condition. Does that
make sense? -Don
On Tue, Jul 27, 2010 at 2:55 PM, Gary King <king(a)harvard.edu> wrote:
both of these would 'work', and both
have been done before lots of times I
think. The only issue is that Amelia isn't 'aware' of the structural
zeros
in the second case or the interactions in the first. e.g., for the
interactions, if you include A, B, and A*B as variables, it doesn't know
that the product of the first 2 variables is the 3rd and so the imputations
won't necessarily respect that known equality. The way we usually deal with
that is to do the imputation unconstrained with all 3 variables and then fix
it after the fact, such as by discarding A*B and multiplying the first two
columns together. This will usually give pretty reasonable results,
although there may be a new method someone could devise (and program!) that
could improve on this approach.
Gary
---
http://gking.harvard.edu
On Tue, Jul 27, 2010 at 12:14 PM, Donald Braman <donald.braman(a)gmail.com>wrote;wrote:
--
Donald Braman
phone: 971-645-0607
http://www.culturalcognition.net/braman/
http://ssrn.com/author=286206
http://www.law.gwu.edu/Faculty/profile.aspx?id=10123
I have a question for the MI experts about
imputations and experiments:
We often run experiments in which we hypothesize that responses will vary
across conditions. We expose subjects to a condition -- CONDa, CONDb, or
CONDc, say -- and then measure responses to a dependent variables across all
conditions, say DV1, DV2, etc. We also collect data on various independent
variables, say IV1, IV2, etc.
But because we anticipate the relationship between the IVs and DVs to
vary across the conditions it seems like we ought to do one of two things
when imputing missing data:
(1) interact every IV with the conditions so that we have, in effect
DV1a, DV1_CONDb, DV1_CONDc, DV2_CONDa, etc. But quite often, the result of
that equation will just result in a zero when the dummy for that condition
is zero rather than one. This seems wasteful of information to me. Which
leads me to my alternative...
(2) Rather than computing DV_CONDa as equalling zero when CONDa is also
zero, I'm tempted to treat every alternative DV (DV1_CONDa, DV1_CONDb,
DV1_CONDc ...) as missing for each case and impute it. Missingness will be
high, of course (only 1/conditions of DVs will be present), but at least I
won't be throwing away lots of valuable data. I hesitate to do so because
I can't find anyone else who has done this and that makes me think I am
probably misguided.
ps. I realize that this is a question about imputation generally, but I
thought I'd post it here since I use Amelia for my imputation needs -- let
me know if I should not post something like this here & I'll look elsewhere
Donald Braman
phone: 413-628-1221
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia
More info about Amelia: http://gking.harvard.edu/amelia
5042
days inactive
5044
days old
amelia@lists.gking.harvard.edu
3 comments
3 participants
participants (3)
-
Donald Braman -
Donald Braman -
Gary King