Previous studies have documented that smoking during pregnancy (SDP) is associated with offspring externalizing problems, even when
measured covariates were used to control for possible confounds.
The adjusted means (controlling for
measured covariates) are based on model 3.
Model 2 was fit to the relation between PAE and offspring externalizing behaviors controlling for
measured covariates that could confound the association (listed in the «Measures» subsection of the «Methods» section).
Ultimately, an environmentally mediated causal effect of PAE on childhood externalizing behaviors — a causal inference — would be supported if the relation remained robust to
the measured covariates and quasi-experimental methods because these approaches account for many of the alternative explanations for the association between PAE and externalizing problems.
Not exact matches
Using age - and gender - specific z scores for the repeated
measures outcome variable and birth weight as the
covariate, the between - group variable (breastfeeding medication group) was significant (P =.005).
This method assumes that the exposures and outcome
measures are missing completely at random given the observed variables and the imputed
covariates.
Associations Between Duration of Breastfeeding and
Measures of Cognitive Ability, Teacher Ratings of School Performance, Standardized Tests of Achievement, and High School Success After Adjustment for
Covariates
Baseline variables were
measured using established valid instruments and were used as
covariates to adjust for differences between randomisation groups in some of the analyses in the paper.
These results suggest that the effect of PMI on
measured gene expression is relatively modest and can be further minimized by using appropriate
covariate correction in analyses.
Therefore we decided to create residual scores for all the relevant spring reading
measures, using appropriate fall scores as
covariates.
The relationship remained significant when per capita Gross Domestic Product was entered as a
covariate as well as when latitude, a
measure of historical climate as well as ultraviolet radiation exposure (Hancock et al., 2008), was controlled for.
A
covariate was included in the multivariate analyses if theoretical or empirical evidence supported its role as a risk factor for obesity, if it was a significant predictor of obesity in univariate regression models, or if including it in the full multivariate model led to a 5 % or greater change in the OR.48 Model 1 includes maternal IPV exposure, race / ethnicity (black, white, Hispanic, other / unknown), child sex (male, female), maternal age (20 - 25, 26 - 28, 29 - 33, 34 - 50 years), maternal education (less than high school, high school graduation, beyond high school), maternal nativity (US born, yes or no), child age in months, relationship with father (yes or no), maternal smoking during pregnancy (yes or no), maternal depression (as
measured by a CIDI - SF cutoff score ≥ 0.5), maternal BMI (normal / underweight, overweight, obese), low birth weight (< 2500 g, ≥ 2500 g), whether the child takes a bottle to bed at age 3 years (yes or no), and average hours of child television viewing per day at age 3 years (< 2 h / d, ≥ 2 h / d).
Analyses used an analysis of covariance (continuous
measure) or logistic regression (categorical
measure) with the posttreatment
measure as a
covariate.
These models pooled the repeated
measures on each outcome at ages 18 to 21 years and 21 to 25 years to produce an estimate of the population - averaged effect of the level of depression at ages 17 to 18 years on the outcome after adjustment for
covariates.
It has been shown that inferences resulting from this analysis are virtually identical no matter which of these outcome
measures is used.30 In addition to the
covariates previously noted, the regression analysis was repeated to include annual household income, mother's treatment setting (primary vs psychiatric outpatient care), and treatment status of child during the 3 - month follow - up period in order to investigate the further potential confounding effects of these variables.
[The moderated regression models were retested using two health - related variables as additional
covariates on Step 1, scores on a standard single item
measuring self - rated health and a count of 17 health conditions for which the respondent had received physician diagnoses (e.g., arthritis, ulcers, emphysema).
It also included the
covariates maternal age and educational level at registration and level of domestic violence in the family (number of incidents of violence)
measured over the 15 - year study period.
All
covariates were
measured at registration and tested for homogeneity of regressions for the hypothesized contrasts.19
There were significant associations between the
covariate measures and both television viewing and educational achievement.
Because of substantial missing data on 2 direct parenting
measures (29 %), multiple imputation via chained equations was used to handle missing
covariate data.30 This approach uses regression models to predict missing data from available variables with 20 imputation iterations selected.
3 The effects shown here are statistically adjusted (using regression
covariates) to account for any pre-program differences between counties in the outcome
measure over the five - year period prior to intervention delivery, as reported in Prinz et al. 2016.
Covariates included age, gender, ethnicity, and
measures of baseline problem behaviors.
Continuous
measures were evaluated using end - point analysis with baseline score as
covariates in both modified intention - to - treat and completer analyses for the self - reported
measures, ICG, Work and Social Adjustment Scale, and Beck Depression and Anxiety Inventory scales.
Table 2 contains the GLM and logistic regressions assessing the contribution of the independent variables, CU levels, and the presence / absence of ODD on the children's psychological
measures for the total sample (n = 622), adjusted by the
covariates family SES, children's ethnicity and sex, other comorbid disorder different from ODD and the number of DSM - IV CD symptoms.
In addition, since from a practical - clinical perspective effect sizes are the most relevant objective of the analyses, and due to the fact that p - values are strongly dependent on sample size, all effect sizes for the relationships analyzed have been estimated by the confidence interval for the parameters, with the R2
measuring the global predictive capacity of the models (adjusted to the
covariates).
To adjust family outcomes at each point for the three concurrent
measures of the family environment, the model also included these factors as time - varying
covariates.
This allowed us to test within subjects experimental effects, considering attachment
measures as
covariates, and to perform repeated
measure analyses, including those participants who were evaluated in only two conditions.
For this modeling, the
measures of CU (ICU - total raw score) and ODD (binary diagnosis present / absent) were considered as the independent variables and the analyses were adjusted by the
covariates family SES, children's sex and ethnicity, presence of comorbidities other than ODD and the number of DSM - IV CD symptoms.
These
covariates were; age, antisocial behaviour severity, autism spectrum disorder (ASD) severity, attention deficit hyperactivity disorder (ADHD) severity, anxiety / depression severity and the Quality of the Family Environment (QFE)
measure.
Research also needs to adequately control for
covariates that may confound the effects of PAE, such as family processes (eg, problematic parenting or family conflict) and parental characteristics, especially maternal substance use.1, 12 Researchers also need to account for genetic liabilities that are shared by parents and offspring.13, 14 A woman's genetic risk of substance use could be passed down to her children and subsequently affect their behavior.15 Research on the consequences of PAE, therefore, needs studies with large samples, with sufficient statistical power to detect small effects, using analytical methods and designs that can account for potential confounds, including factors that are not
measured.
Additionally, the results of the GLM, adjusted by the
covariates of the study,
measuring the association of the CU raw score on the ODD
measures also at age 3 for the children diagnosed with ODD at baseline (n = 61) showed that CU levels did not achieved significant contribution on the ODD level (p ≥.810; R2 ≤.006).
Therefore, we included these
measures as
covariates in the regression models.
Post-conceptual age on the visit day of EEG was entered as
covariate for the EEG
measures.
First, we examined the relationships of plausible
covariates, including gender, birth - weight, post-conceptual age on the visit day (gestational age + days of life since birth to the visit day), ethnicity, prenatal smoking exposure, and child sleep condition at the time of EEG recording with outcome
measures (frontal EEG power, functional connectivity at 6 and 18 months of age, or behavioral scores at 24 months of age).
Thus verbal IQ (T1,
measured concurrently with the MCAST) was entered as
covariate in further analysis regarding MCAST attachment to mother.
The same study design and
measures allowed us to adjust for the
covariates of size and IQ.