Debates about our design are beside the point: The Reardon and Cougle findings are invalid and cannot be reproduced with properly coded data.
The rapid
responses to our article[1] have raised a variety of
issues. Here we address the methodological issues that have been raised by
David Reardon [2] and others, as noted.To
avoid redundancy, theoretical, ethical, and clinical issues raised across the
responses will be addressed in subsequent responses.
In his response, Reardon
takes the opportunity to repeat the results presented in his original article (co-authored
with Jesse Cougle), arguing that we have “failed to
present the most basic evidence necessary to either refute or confirm our prior
results.”He states “their results do
not contradict ours; indeed, they do not even attempt to reconstruct our
analysis in regard to stratification by marital status using their recoded
coded data. Even what they do report… can easily be reconciled with our own
findings since any differences in results ‘can primarily be explained by
differences in coding of key variables and sample selection’.”
1. Results based on miscoded data are not in need of refuting and
cannot be reconciled.
In our attempt to take
a collegial approach, we have perhaps not been clear about the basic
methodological message of our article. The previous results were based on miscoded
data such that first unintended pregnancies and their outcomes (pregnancy
versus abortion) were not accurately identified. We are not talking
about a difference in interpretation of how best to code a particular variable,
e.g., as when researchers need to decide whether to use all possible categories
for marital status or to create a variable that only has 2 categories (i.e.,
married and unmarried). We are talking about the use of codes to identify the
variable the former authors themselves claimed to have measured. Results
based on miscoded data are not in need of confirmation or refutation, as
results based on such data are invalid and meaningless.
As stated in our
article, the NLSY dataset is large and complex, with over 3,000 variables
related to pregnancy. As can be seen by the coding syntax appended to this
response, identification of first pregnancy is particularly complex, as it
involves combining variables across multiple survey years. Even the most
competent researcher may have difficulty in writing codes that accurately specify
the variable “first unintended pregnancy”. Because of the enormous room for error in
choosing the proper variables and coding of variables in this large dataset, we
chose to rely on the variable selection strategy and coding language provided
to us by an expert in the use of the survey: a member of the NLSY staff.
Reardon suggests that we
should make our codes available so that our findings can be checked. The codes,
which are in SAS, can be found at the end of this response. We are confident
that the codes produce valid results, i.e., that they create the variable that
we are claiming to measure. The coding was provided to us by NLSY staff and the
logic was rechecked several times. However, if there is a problem with this
coding language, we welcome learning about it. As embarrassing as that would
be, our goal is to produce accurate research findings. We welcome constructive
feedback on how to achieve that goal.
2.There was no need to
reproduce previously reported tables based on results from analyses that we
considered inappropriate for the research question.
Our study was
designed to test the hypothesis that was offered (but tested with miscoded
data) in the earlier study.In doing so,
we found several design assumptions that we did not consider appropriate for
testing that hypothesis.We therefore did
not incorporate them in our study, but did explain the rationale for our
decisions.We also noted that we indeed
did conduct analyses to parallel those of the previous study by examining
pregnancies limited to 1980 and later, but did not find significant results. Typically
journals have limited space and are not receptive to publishing nonsignificant findings that are based on inappropriate
analyses, so we did not include them in our paper. Nonetheless, we are happy to present these
analyses here and to elaborate the rationale for our decisions.
Most of the design
issues relate to the claim that our sample is biased because: (a) we did not
exclude those in either the abortion or delivery group who had subsequent
abortions after the first pregnancy (submitted by Fiona Pinto); (b) we did not
stratify our results by marital status (submitted by David Reardon); (c) we excluded abortions where the
pregnancy was reported as “wanted” (submitted by Patrick Leahy);and (d) the age
range of our sample was limited (submitted by Patrick Leahy).
2a. If the goal is to generalize the results to women having a first
unintended pregnancy, exclusion of women with multiple abortions from either
the delivery group or the abortion group is inappropriate.Exclusions of women with multiple abortions
from only the delivery group are doubly inappropriate.Whether one agrees with this view is
irrelevant, however, because the results
are the same in any case.
Women dealing with
an unwanted first pregnancy need information that can help them assess how
choosing one outcome versus another (i.e., delivery vs. abortion) will
contribute to a change in risk for physical and mental health outcomes. We
could have eliminated women having multiple abortions from both groups, but
without a crystal ball, women who will have multiple abortions cannot be
separated from other women. Consequently, information based on such a sample
has no useful medical purpose.
But regardless of
whether one agrees with our judgment, as we clearly stated on p. 2 of the
manuscript and as can be seen in the following tables, the conclusions did not
differ when women were excluded on the basis of subsequent abortion, regardless
of whether that exclusion was from both the abortion and delivery groups, or
simply the delivery group.
Logistic and OLS regression
results if all women with subsequent abortions are excluded:
2b. Results do not differ by current marital status.
Arguments could be
made about whether to stratify by marital status the year depression was
measured or the year the unwanted pregnancy occurred, but they are irrelevant here because the findings
are consistent across marital status.We
stratified by marital status in an earlier iteration of the paper, but dropped
the table because it didn’t contribute information of significance. The
following table presents the same findings found in our Table 2, but is now stratified
on the basis of “married” versus “unmarried” in 1992.As can be seen, the findings are consistent
across marital groups for both our logistic and OLS regression results.
Updates of Table 2, stratified by marital status in 1992
Delivery Group133/369 (36.0).83 (.58 to 1.18)1.32 (.83 to 2.09)#
Abortion Group 71/223(31.8)P = 0.30P = 0.25
Married Women:
Delivery Group85/393(21.6) .81 (.55 to 1.21)1.08
(.67 to 1.76)#
Abortion Group 46/251(18.3)P = 0.31P
= 0.75
All Women:
Delivery Group220/768 (28.6).82 (.64 to 1.07)1.19 (.85 to 1.66)*
Abortion Group 119/479 (24.8)P = 0.14P = 0.30
Post 1979
Pregnancies only
Unmarried Women:
Delivery Group65/200 (32.5).91 (.55 to
1.51)1.39 (.71 to 2.71)&
Abortion Group 33/108 (30.6)P = 0.73P
= 0.34
Married Women:
Delivery Group45/256 (17.6).98 (.55 to 1.72)1.31 (.69 to 2.49)&
Abortion Group 21/122(17.2)P = 0.93P = 0.42
All Women:
Delivery Group 111/457
(24.3).97 (.67 to 1.40)1.33 (.84 to 2.10)†
Abortion Group55/232 (23.7)P = 0.87P = 0.23
Note. Pregnancy outcome coded "0" for
delivery and "1" for abortion; higher CES-D scores indicate greater
levels of depression.
Odds ratios greater
than 1.0 would thus indicate a higher risk for depression in the abortion group
*Adjusted for race,
age at first pregnancy, and 1992 marital status, education, and family income
# Adjusted
for race, age at first pregnancy, education, and family income
&Adjusted for 1979 measure of Rotter internal locus of
control, race, age at first pregnancy, education, and family income
† Adjusted for 1979
measure of Rotter internal locus of control, race, age at first
pregnancy, and 1992 marital status, education, and family income
OLS Regression predicting CES-D continuous scores
Mean (SD)Unadjusted Adjusted
Beta
(S.E)Beta (S. E)
Full Sample
Unmarried Women:
Delivery Group13.6 (10.4)-.63 (.89)1.55 (1.13)#
Abortion Group 13.0 (10.6)P
= 0.48P = 0.17
Married Women:
Delivery Group10.1 (9.0)-1.31 (.72)-.55 (.82)#
Abortion Group 8.8 (8.7)P
= 0.07P =
0.51
All Women:
Delivery Group11.8 (9.95) -1.07 (.58) .38 (.68)*
Abortion Group10.8 (9.9)P = 0.06 P = 0.58
Post 1979
Pregnancies only
Unmarried Women:
Delivery Group13.0 (10.1).35
(1.28)1.80 (1.63)&
Abortion Group 13.4 (11.7)P
= 0.78P = 0.27
Married Women:
Delivery Group8.8 (8.0) -.37 (.90).56
(1.00)&
Abortion Group 8.4 (8.4) P = 0.68P = 0.58
All Women:
Delivery Group10.7 (9.4) .07 (.78).92 (.87)†
Abortion Group10.8 (10.4)P = 0.93P
= 0.29
Note. Pregnancy outcome coded "0" for
delivery and "1" for abortion; higher CES-D scores indicate greater
levels of depression.
Positive beta
coefficients would thus indicate a higher risk for depression in the abortion group.
*Adjusted for race,
age at first pregnancy, and 1992 marital status, education, and family income
# Adjusted
for race, age at first pregnancy, education, and family income
&Adjusted for 1979 measure of Rotter internal locus of
control, race, age at first pregnancy, education, and family income
† Adjusted for 1979
measure of Rotter internal locus of control, race, age at first
pregnancy, and 1992 marital status, education, and family income
2c.In a study of unwanted pregnancy outcome, it
is inappropriate to include wanted pregnancies in the sample; it is doubly
inappropriate to exclude wanted pregnancies from only the delivery outcome
group.The stated purpose of our study
was to examine the relationship between pregnancy outcome and depression of unwanted first pregnancies.We chose this as our research question and
designed our analyses accordingly.
But
regardless of one’s opinion about the appropriateness of the particular design strategy,
as we state in the paper only 15 women (of a total sample of n=1247) who terminated
their first pregnancy by abortion reported that pregnancy as wanted. Although
we excluded these women because of design considerations, redoing the analyses
with these 15 women included produces no change in the results.
2d. The age range of our sample was not limited to 14-21 years of age.Our description of the age of the sample appears
to be misinterpreted by Patrick Leahy. The age range of women was 14-21 in
1979, the year the study began. The year of first pregnancy ranged from 1970 to
1992, and the age of women at the first pregnancy ranged from 12-33.
In sum, there may
be debates about our design decisions, which we defend.However, the most important point here is
that these design issues do not affect the outcome of the results when data are
properly coded.
3. The critique of our paper, as well as the public discussion of research
on the relationship of abortion and mental health outcomes, has exhibited a persistent
failure to distinguish between correlation and causation in the interpretation
of results.
This failure is
clearly illuminated in Reardon’s statement: “First, it should be noted that the authors’ claim…stating “Well
designed studies have not found that abortion contributes to an increased risk
of depression,” is misleading and not supported by their discussion, their
citations, or the literature. In fact,
the statistical association between abortion and higher depression rates is
very firmly established by many well designed studies.”[italics ours] “
Reardon then cites a number of studies, some more
well designed than others, but none that have established that abortion
contributes to an increased risk of depression.Indeed, he goes on to describe an article by
Russo and Denious [3}, repeating their findings that women who had abortions
had significantly more depression, suicidal ideation, and lower life
satisfaction than other women. He then describes
those researchers as arguing that “this association with depression might
possibly be explained by greater exposure to experiences of violence among
women who have abortion” and asserts “the act of simply proposing this
hypothesis serves to demonstrate the fact that the irrefutable evidence of a
link between abortion and depression requires explanation and further
investigation.” What he fails to say, is that Russo and Denious did not “simply propose” this hypothesis.They tested it and found that when
exposure to violence and partner variables were controlled, no relationship was
found between abortion and the negative mental health outcomes measured.
Reardon then goes
on to cite as evidence that abortion causes depression “self-attributions of
women,” “clinical experience of counselors,” “case studies,”
and statistical evidence based on studies that have not included the basic
controls needed to warrant causal conclusions.None of this is credible scientific evidence
that abortion increases risk for depression.
Many studies, including our own, have found correlations
between abortion and a host of mental health outcomes. Given that abortion
typically occurs in the context of unwanted pregnancy, this association is to
be expected. The error is in focusing on abortion rather than the conditions
that lead to risk for unplanned and unwanted pregnancy. As Russo and Denious pointed
out, having a history of childhood sexual abuse and exposure to intimate
violence is associated with risk for unplanned and unwanted pregnancy, whether
or not such pregnancy ends in abortion or delivery.In this context, focusing on the effects of
abortion rather than trying to understand the relationship of abortion to
mental health outcomes can lead to misattribution of the effects of childhood
physical and sexual abuse, intimate partner violence, and other adversities.
We will deal with issues
related to discrepancies between scientific, clinical, and anecdotal evidence
raised by the various responses as well as in issues relating to media coverage
and future research needs, including the need for research on the effects of
underreporting, in separate replies.
For now however,
we stand by the statement that there is no crediblescientific evidence that abortion increases risk of depression.
[1]Schmiege S, Russo NF.
Depression and unwanted first pregnancy: longitudinal cohort study.
[3]Russo N, Denious JE. Violence in the lives of
women having abortions: Implications for policy and practice. Professional Psychology Research and Practice, 2001); 32:142-150.
SAS SYNTAX
libname
in 'g:\british';
OPTIONS
SYSPARM='.';
OPTIONS
NOCENTER;
OPTIONS
LS=80;
DATA
a; SET in.nlsy_1110;
rename R1017500=wantp183;
rename R1325400=wantp184;
rename R1522056=outp1_84;
rename R1702800=wantp185;
rename R1892758=outp1_85;
rename R2002900=wantp186;
rename R2259858=outp1_86;
rename R2579200=wantp188;
rename R2879700=outp1_88;
rename R3188900=wantp190;
rename R3409800=outp1_90;
rename R3792500=wantp192;
rename R4009468=outp1_92;
data b; set
a;
if R0214800=2;
procfreq;
tables wantp183 wantp184
outp1_84
wantp185
outp1_85
wantp186
outp1_86
wantp188
outp1_88
wantp190
outp1_90
wantp192
outp1_92;
data c; set
b;
p1_out=.;
if outp1_92>-4then p1_out=outp1_92;
if outp1_90>-4then p1_out=outp1_90;
if outp1_88>-4then p1_out=outp1_88;
if outp1_86>-4then p1_out=outp1_86;
if outp1_85>-4then p1_out=outp1_85;
if outp1_84>-4then p1_out=outp1_84;
wanted=.;
if p1_out>-4
and p1_out6thendo;
if outp1_84>-4
and wantp183>=-3then
wanted=wantp183;
elseif
outp1_84>-4 and wantp184>=-3then wanted=wantp184;
elseif
outp1_85>-4 and wantp185>=-3then wanted=wantp185;
elseif
outp1_86>-4 and wantp186>=-3then wanted=wantp186;
elseif
outp1_88>-4 and wantp188>=-3then wanted=wantp188;
elseif
outp1_90>-4 and wantp190>=-3then wanted=wantp190;
elseif
outp1_92>-4 and wantp192>=-3then wanted=wantp192;
elseif
wanted=.and
outp1_92 in (2345) then wanted=999;
end;
data d; set
c;
if wanted=3
or wanted=4thendo;
if p1_out = 1thenpregout = 0;
end;
if wanted ne1thendo;
if p1_out=4thenpregout = 1;
end;
datacesd; set d;
x
= n (R3894900, R3895000, R3895100, R3895200, R3895300, R3895400, R3895500,
Rapid Response:
Debates about our design are beside the point: The Reardon and Cougle findings are invalid and cannot be reproduced with properly coded data.
The rapid
responses to our article[1] have raised a variety of
issues. Here we address the methodological issues that have been raised by
David Reardon [2] and others, as noted. To
avoid redundancy, theoretical, ethical, and clinical issues raised across the
responses will be addressed in subsequent responses.
In his response, Reardon
takes the opportunity to repeat the results presented in his original article (co-authored
with Jesse Cougle), arguing that we have “failed to
present the most basic evidence necessary to either refute or confirm our prior
results.” He states “their results do
not contradict ours; indeed, they do not even attempt to reconstruct our
analysis in regard to stratification by marital status using their recoded
coded data. Even what they do report… can easily be reconciled with our own
findings since any differences in results ‘can primarily be explained by
differences in coding of key variables and sample selection’.”
1. Results based on miscoded data are not in need of refuting and
cannot be reconciled.
In our attempt to take
a collegial approach, we have perhaps not been clear about the basic
methodological message of our article. The previous results were based on miscoded
data such that first unintended pregnancies and their outcomes (pregnancy
versus abortion) were not accurately identified. We are not talking
about a difference in interpretation of how best to code a particular variable,
e.g., as when researchers need to decide whether to use all possible categories
for marital status or to create a variable that only has 2 categories (i.e.,
married and unmarried). We are talking about the use of codes to identify the
variable the former authors themselves claimed to have measured. Results
based on miscoded data are not in need of confirmation or refutation, as
results based on such data are invalid and meaningless.
As stated in our
article, the NLSY dataset is large and complex, with over 3,000 variables
related to pregnancy. As can be seen by the coding syntax appended to this
response, identification of first pregnancy is particularly complex, as it
involves combining variables across multiple survey years. Even the most
competent researcher may have difficulty in writing codes that accurately specify
the variable “first unintended pregnancy”. Because of the enormous room for error in
choosing the proper variables and coding of variables in this large dataset, we
chose to rely on the variable selection strategy and coding language provided
to us by an expert in the use of the survey: a member of the NLSY staff.
Reardon suggests that we
should make our codes available so that our findings can be checked. The codes,
which are in SAS, can be found at the end of this response. We are confident
that the codes produce valid results, i.e., that they create the variable that
we are claiming to measure. The coding was provided to us by NLSY staff and the
logic was rechecked several times. However, if there is a problem with this
coding language, we welcome learning about it. As embarrassing as that would
be, our goal is to produce accurate research findings. We welcome constructive
feedback on how to achieve that goal.
2. There was no need to
reproduce previously reported tables based on results from analyses that we
considered inappropriate for the research question.
Our study was
designed to test the hypothesis that was offered (but tested with miscoded
data) in the earlier study. In doing so,
we found several design assumptions that we did not consider appropriate for
testing that hypothesis. We therefore did
not incorporate them in our study, but did explain the rationale for our
decisions. We also noted that we indeed
did conduct analyses to parallel those of the previous study by examining
pregnancies limited to 1980 and later, but did not find significant results. Typically
journals have limited space and are not receptive to publishing nonsignificant findings that are based on inappropriate
analyses, so we did not include them in our paper. Nonetheless, we are happy to present these
analyses here and to elaborate the rationale for our decisions.
Most of the design
issues relate to the claim that our sample is biased because: (a) we did not
exclude those in either the abortion or delivery group who had subsequent
abortions after the first pregnancy (submitted by Fiona Pinto); (b) we did not
stratify our results by marital status
(submitted by David Reardon); (c) we excluded abortions where the
pregnancy was reported as “wanted” (submitted by Patrick Leahy);and (d) the age
range of our sample was limited (submitted by Patrick Leahy).
2a. If the goal is to generalize the results to women having a first
unintended pregnancy, exclusion of women with multiple abortions from either
the delivery group or the abortion group is inappropriate. Exclusions of women with multiple abortions
from only the delivery group are doubly inappropriate. Whether one agrees with this view is
irrelevant, however, because the results
are the same in any case.
Women dealing with
an unwanted first pregnancy need information that can help them assess how
choosing one outcome versus another (i.e., delivery vs. abortion) will
contribute to a change in risk for physical and mental health outcomes. We
could have eliminated women having multiple abortions from both groups, but
without a crystal ball, women who will have multiple abortions cannot be
separated from other women. Consequently, information based on such a sample
has no useful medical purpose.
But regardless of
whether one agrees with our judgment, as we clearly stated on p. 2 of the
manuscript and as can be seen in the following tables, the conclusions did not
differ when women were excluded on the basis of subsequent abortion, regardless
of whether that exclusion was from both the abortion and delivery groups, or
simply the delivery group.
Logistic and OLS regression
results if all women with subsequent abortions are excluded:
No. (%) exceeding Unadjusted Adjusted
CES-D
cut-off OR (95% CI) OR
(95% CI)
Dichotomous CESD:
Delivery Group 185/646
(28.6) .66 (.48 to .90) 1.14 (.76 to 1.70)*
Abortion Group 67/322 (20.8)
P
= 0.009 P
= 0.54
Mean (SD) Unadjusted Adjusted
Beta
(S.E) Beta
(S. E)
Continuous CESD:
Delivery Group 11.8 (9.9) -2.0
(.65) .36 (.78)*
Abortion Group 9.8 (9.1) P =
0.002 P =
0.64
*Adjusted for race, age at first
pregnancy, and 1992 marital status, education, and family income
Logistic
and OLS regression results if only women in the delivery group with subsequent abortions
are excluded:
No. (%) exceeding Unadjusted Adjusted
CES-D
cut-off OR (95% CI) OR
(95% CI)
Dichotomous CESD:
Delivery Group 185/646 (28.6)
.82 (.63 to 1.08) 1.17 (.82
to 1.65)*
Abortion Group 119/479 (24.8) P = 0.16 P = 0.39
Mean (SD) Unadjusted Adjusted
Beta
(S.E) Beta
(S. E)
Continuous CESD:
Delivery Group 11.8 (9.9) -1.01
(.60) .26 (.70)*
Abortion Group 10.8 (9.9) P = 0.09 P =
0.71
*Adjusted
for race, age at first pregnancy, and 1992 marital status, education,
and family income
2b. Results do not differ by current marital status.
Arguments could be
made about whether to stratify by marital status the year depression was
measured or the year the unwanted pregnancy occurred, but they are irrelevant here because the findings
are consistent across marital status. We
stratified by marital status in an earlier iteration of the paper, but dropped
the table because it didn’t contribute information of significance. The
following table presents the same findings found in our Table 2, but is now stratified
on the basis of “married” versus “unmarried” in 1992. As can be seen, the findings are consistent
across marital groups for both our logistic and OLS regression results.
Updates of Table 2, stratified by marital status in 1992
Logistic Regression predicting CES-D
cut-off scores
No.
(%) exceeding Unadjusted Adjusted
CES-D
cut-off OR
(95% CI) OR
(95% CI)
Full Sample
Unmarried Women:
Delivery Group 133/369 (36.0) .83 (.58 to 1.18) 1.32 (.83 to 2.09)#
Abortion Group 71/223 (31.8) P = 0.30 P = 0.25
Married Women:
Delivery Group 85/393 (21.6)
.81 (.55 to 1.21) 1.08
(.67 to 1.76)#
Abortion Group 46/251 (18.3) P = 0.31 P
= 0.75
All Women:
Delivery Group 220/768 (28.6) .82 (.64 to 1.07) 1.19 (.85 to 1.66)*
Abortion Group 119/479 (24.8) P = 0.14 P = 0.30
Post 1979
Pregnancies only
Unmarried Women:
Delivery Group 65/200 (32.5) .91 (.55 to
1.51) 1.39 (.71 to 2.71)&
Abortion Group 33/108 (30.6) P = 0.73 P
= 0.34
Married Women:
Delivery Group 45/256
(17.6) .98 (.55 to 1.72) 1.31 (.69 to 2.49)&
Abortion Group 21/122 (17.2) P = 0.93 P = 0.42
All Women:
Delivery Group 111/457
(24.3) .97 (.67 to 1.40) 1.33 (.84 to 2.10)†
Abortion Group 55/232 (23.7) P = 0.87 P = 0.23
Note. Pregnancy outcome coded "0" for
delivery and "1" for abortion; higher CES-D scores indicate greater
levels of depression.
Odds ratios greater
than 1.0 would thus indicate a higher risk for depression in the abortion group
*Adjusted for race,
age at first pregnancy, and 1992 marital status, education, and family income
# Adjusted
for race, age at first pregnancy, education, and family income
&Adjusted for 1979 measure of Rotter internal locus of
control, race, age at first pregnancy, education, and family income
† Adjusted for 1979
measure of Rotter internal locus of control, race, age at first
pregnancy, and 1992 marital status, education, and family income
OLS Regression predicting CES-D continuous scores
Mean (SD) Unadjusted Adjusted
Beta
(S.E) Beta (S. E)
Full Sample
Unmarried Women:
Delivery Group 13.6 (10.4) -.63 (.89) 1.55 (1.13)#
Abortion Group 13.0 (10.6) P
= 0.48 P = 0.17
Married Women:
Delivery Group 10.1 (9.0) -1.31 (.72) -.55 (.82)#
Abortion Group 8.8 (8.7) P
= 0.07 P =
0.51
All Women:
Delivery Group 11.8 (9.95)
-1.07 (.58) .38 (.68)*
Abortion Group 10.8 (9.9) P = 0.06 P = 0.58
Post 1979
Pregnancies only
Unmarried Women:
Delivery Group 13.0 (10.1) .35
(1.28) 1.80 (1.63)&
Abortion Group 13.4 (11.7) P
= 0.78 P = 0.27
Married Women:
Delivery Group 8.8 (8.0)
-.37 (.90) .56
(1.00)&
Abortion Group 8.4 (8.4)
P = 0.68 P = 0.58
All Women:
Delivery Group 10.7 (9.4) .07 (.78) .92 (.87)†
Abortion Group 10.8 (10.4) P = 0.93 P
= 0.29
Note. Pregnancy outcome coded "0" for
delivery and "1" for abortion; higher CES-D scores indicate greater
levels of depression.
Positive beta
coefficients would thus indicate a higher risk for depression in the abortion group.
*Adjusted for race,
age at first pregnancy, and 1992 marital status, education, and family income
# Adjusted
for race, age at first pregnancy, education, and family income
&Adjusted for 1979 measure of Rotter internal locus of
control, race, age at first pregnancy, education, and family income
† Adjusted for 1979
measure of Rotter internal locus of control, race, age at first
pregnancy, and 1992 marital status, education, and family income
2c. In a study of unwanted pregnancy outcome, it
is inappropriate to include wanted pregnancies in the sample; it is doubly
inappropriate to exclude wanted pregnancies from only the delivery outcome
group. The stated purpose of our study
was to examine the relationship between pregnancy outcome and depression of unwanted first pregnancies. We chose this as our research question and
designed our analyses accordingly.
But
regardless of one’s opinion about the appropriateness of the particular design strategy,
as we state in the paper only 15 women (of a total sample of n=1247) who terminated
their first pregnancy by abortion reported that pregnancy as wanted. Although
we excluded these women because of design considerations, redoing the analyses
with these 15 women included produces no change in the results.
2d. The age range of our sample was not limited to 14-21 years of age. Our description of the age of the sample appears
to be misinterpreted by Patrick Leahy. The age range of women was 14-21 in
1979, the year the study began. The year of first pregnancy ranged from 1970 to
1992, and the age of women at the first pregnancy ranged from 12-33.
In sum, there may
be debates about our design decisions, which we defend. However, the most important point here is
that these design issues do not affect the outcome of the results when data are
properly coded.
3. The critique of our paper, as well as the public discussion of research
on the relationship of abortion and mental health outcomes, has exhibited a persistent
failure to distinguish between correlation and causation in the interpretation
of results.
This failure is
clearly illuminated in Reardon’s statement:
“First, it should be noted that the authors’ claim…stating “Well
designed studies have not found that abortion contributes to an increased risk
of depression,” is misleading and not supported by their discussion, their
citations, or the literature. In fact,
the statistical association between abortion and higher depression rates is
very firmly established by many well designed studies.”[italics ours] “
Reardon then cites a number of studies, some more
well designed than others, but none that have established that abortion
contributes to an increased risk of depression. Indeed, he goes on to describe an article by
Russo and Denious [3}, repeating their findings that women who had abortions
had significantly more depression, suicidal ideation, and lower life
satisfaction than other women. He then describes
those researchers as arguing that “this association with depression might
possibly be explained by greater exposure to experiences of violence among
women who have abortion” and asserts “the act of simply proposing this
hypothesis serves to demonstrate the fact that the irrefutable evidence of a
link between abortion and depression requires explanation and further
investigation.” What he fails to say, is that Russo and Denious did not “simply propose” this hypothesis. They tested it and found that when
exposure to violence and partner variables were controlled, no relationship was
found between abortion and the negative mental health outcomes measured.
Reardon then goes
on to cite as evidence that abortion causes depression “self-attributions of
women,” “clinical experience of counselors,” “case studies,”
and statistical evidence based on studies that have not included the basic
controls needed to warrant causal conclusions. None of this is credible scientific evidence
that abortion increases risk for depression.
Many studies, including our own, have found correlations
between abortion and a host of mental health outcomes. Given that abortion
typically occurs in the context of unwanted pregnancy, this association is to
be expected. The error is in focusing on abortion rather than the conditions
that lead to risk for unplanned and unwanted pregnancy. As Russo and Denious pointed
out, having a history of childhood sexual abuse and exposure to intimate
violence is associated with risk for unplanned and unwanted pregnancy, whether
or not such pregnancy ends in abortion or delivery. In this context, focusing on the effects of
abortion rather than trying to understand the relationship of abortion to
mental health outcomes can lead to misattribution of the effects of childhood
physical and sexual abuse, intimate partner violence, and other adversities.
We will deal with issues
related to discrepancies between scientific, clinical, and anecdotal evidence
raised by the various responses as well as in issues relating to media coverage
and future research needs, including the need for research on the effects of
underreporting, in separate replies.
For now however,
we stand by the statement that there is no crediblescientific evidence that abortion increases risk of depression.
[1]Schmiege S, Russo NF.
Depression and unwanted first pregnancy: longitudinal cohort study.
[2] Reardon DC,
Cougle JR. Depression and unintended pregnancy in the
National Longitudinal Survey of Youth: a cohort study. British
Medical Journal.2002; 324:151-2.http://bmj.bmjjournals.com/cgi/reprint/324/7330/151
[3] Russo N, Denious JE. Violence in the lives of
women having abortions: Implications for policy and practice. Professional Psychology Research and Practice, 2001); 32:142-150.
SAS SYNTAX
libname
in 'g:\british';
OPTIONS
SYSPARM='.';
OPTIONS
NOCENTER;
OPTIONS
LS=80;
DATA
a; SET in.nlsy_1110;
rename R1017500=wantp183;
rename R1325400=wantp184;
rename R1522056=outp1_84;
rename R1702800=wantp185;
rename R1892758=outp1_85;
rename R2002900=wantp186;
rename R2259858=outp1_86;
rename R2579200=wantp188;
rename R2879700=outp1_88;
rename R3188900=wantp190;
rename R3409800=outp1_90;
rename R3792500=wantp192;
rename R4009468=outp1_92;
data b; set
a;
if R0214800=2;
procfreq;
tables wantp183 wantp184
outp1_84
wantp185
outp1_85
wantp186
outp1_86
wantp188
outp1_88
wantp190
outp1_90
wantp192
outp1_92;
data c; set
b;
p1_out=.;
if outp1_92>-4then p1_out=outp1_92;
if outp1_90>-4then p1_out=outp1_90;
if outp1_88>-4then p1_out=outp1_88;
if outp1_86>-4then p1_out=outp1_86;
if outp1_85>-4then p1_out=outp1_85;
if outp1_84>-4then p1_out=outp1_84;
wanted=.;
if p1_out>-4
and p1_out6thendo;
if outp1_84>-4
and wantp183>=-3then
wanted=wantp183;
elseif
outp1_84>-4 and wantp184>=-3then wanted=wantp184;
elseif
outp1_85>-4 and wantp185>=-3then wanted=wantp185;
elseif
outp1_86>-4 and wantp186>=-3then wanted=wantp186;
elseif
outp1_88>-4 and wantp188>=-3then wanted=wantp188;
elseif
outp1_90>-4 and wantp190>=-3then wanted=wantp190;
elseif
outp1_92>-4 and wantp192>=-3then wanted=wantp192;
elseif
wanted=.and
outp1_92 in (2345)
then wanted=999;
end;
data d; set
c;
if wanted=3
or wanted=4thendo;
if p1_out = 1thenpregout = 0;
end;
if wanted ne1thendo;
if p1_out=4thenpregout = 1;
end;
datacesd;
set d;
x
= n (R3894900, R3895000, R3895100, R3895200, R3895300, R3895400, R3895500,
R3895600,
R3895700, R3895800, R3895900, R3896000, R3896100, R3896200, R3896300, R3896400,
R3896500,
R3896600, R3896700, R3896800);
if x = 20thencesdcont = sum
(R3894900, R3895000, R3895100, R3895200, R3895300, R3895400, R3895500,
R3895600,
R3895700, R3895800, R3895900, R3896000, R3896100, R3896200, R3896300, R3896400,
R3896500,
R3896600, R3896700, R3896800);
ifcesdcont>=0thendo;
cesddich=.;
ifcesdcont>15thencesddich=1;
elseif
(cesdcont> 0)
or (cesdcont16)
thencesddich=0;
end;
procfreq;
tablespregoutcesddichpregout*cesddich;
run;
procformat;
value p1_out
1='live
birth'
2='miscarriage'
3='stillborn'
4='abortion'
5='still
pregnant';
valuepregout
0='live
birth'
1='abortion';
value wanted
1='wanted
to be preg'
2="didn't
matte"
3='no,
not then'
4='no,
never'
999='not
asked, 1992';
valuecesddich
0='below
cutoff'
1='above
cutoff';
run;
Competing interests:
None declared
Competing interests: The rapidresponses to our article