Objectives. To estimate the effects of maternal and birth characteristics on the decision to breastfeed and to relate breastfeeding practices to racial differences in infant mortality.
Methods. Using a sample of women with young children from the National Survey of Family Growth (NSFG), Cycle V, 1995, the likelihood of breastfeeding was modeled using logistic regression techniques. In addition, single, live births from the NSFG 1988 and 1995 surveys were analyzed to model the effects of race and breastfeeding on infant mortality using Cox regression methods.
Results. After controlling for socioeconomic background and birth characteristics, race remained a strong predictor of breastfeeding. Black women were less likely to breastfeed than nonblack women were, and the primary reason indicated by black women for not breastfeeding was that they “preferred to bottle-feed.” Analyses of infant mortality indicated that breastfeeding accounts for the race difference in infant mortality in the United States at least as well as low birth weight does.
Conclusions. Race is an important predictor of breastfeeding, with most black women reporting that they “preferred bottle-feeding.” Efforts to increase breastfeeding of infants in the black community should help narrow the racial gap in infant mortality.
The importance of breastfeeding to child health in the United States was highlighted by the celebration of World Breastfeeding Week in August 2000. Breastfeeding is the healthiest way for a newborn child to get the best nutrition possible,1 and breast milk provides maternal antibodies that are effective in preventing disease.2,,3 Hence, the slogan “Breast is Best” was created to encourage mothers to breastfeed their children; however, many infants, particularly high-risk infants, do not receive the benefits of breast milk. Given the health benefits of breastfeeding, it is important to know which women have been receptive to the “breast is best” message and which have not.
Previous research has focused mainly on the influence of age, income, and race on the decision to breastfeed.4 Social class is particularly important in predicting whether or not a woman will breastfeed.5 In addition to income, education influences breastfeeding as college educated women are more likely to breastfeed than their less educated counterparts.6–12 Various studies have also found a negative relationship between maternal employment and breastfeeding.6,,7,13 These studies conclude that women who are employed are not as likely to breastfeed as are unemployed women, primarily because of time constraints and job requirements.8,14–16
Demographic factors such as age, race, and ethnicity are also predictive of breastfeeding. Older women are more likely to breastfeed than are younger women,10,,17 and Hispanic mothers are more likely to breastfeed than non-Hispanic mothers.5,,10 In particular, studies have also consistently found racial differences4 in the decision to breastfeed, and foreign born women are more likely to breastfeed than are US-born women.18
Other maternal characteristics such as religion (Catholic women are more likely to breastfeed than are Protestant women10,,15,19) and marital status influence the decision to breastfeed. The support of the infant's father is important in the breastfeeding decision,10 and married women are more likely to breastfeed than are single women.15 Besides partner support, maternal attitudes are also influential. Women with a positive self-image and women who are health conscious are more likely to breastfeed than are their less positive and health conscious counterparts.20,,21 Finally, area of residency and region are also predictive of breastfeeding.11,,15,22
Besides maternal characteristics, studies have found that birth characteristics such as birth order, the type of delivery (vaginal or cesarean), and the health of the infant immediately after delivery influence breastfeeding. First-born children are more likely to be breastfed than are higher parity children15,,23, and infants with poor health are breastfed less than are healthy infants.12,,22 In addition, mothers having a cesarean section are less likely to breastfeed than are mothers who deliver vaginally.24,,25
Thus, various maternal and child characteristics have been found to influence the breastfeeding decision. To examine the combined influence of these characteristics on the likelihood of breastfeeding, we model the effects of maternal and birth factors on the initiation of breastfeeding for a sample of women with children 18 months and younger in 1995. In addition, we examine additional racial differences in the decision to breastfeed and relate racial differences in breastfeeding practices to infant survival.
For this study we analyzed data from the National Survey of Family Growth (NSFG), Cycle V, 1995. These data were collected by the Centers for Disease Control and Prevention and include a national sample of women of childbearing age. To reduce recall error in analyzing factors associated with the breastfeeding decision, we limited our sample to women with a child 18 months or younger at the date of interview (N = 1088). Eighteen months was chosen to have sufficient cases for analysis of racial differences; however, analyses were also performed for women with children 1 year old and younger and the effects, although reduced somewhat because of smaller sample size, were essentially the same (available from authors). The sample includes only single-birth, surviving infants who lived with the mother.
The Breastfeeding Model
Based on this sample of mothers with children 18 months and younger, we examined the effects of various maternal and birth characteristics on the decision to ever breastfeed the child. The variables are outlined in Table 1. The dependent variable indicated whether or not the mother ever breastfed the child.
Maternal characteristics included measures of socioeconomic background, such as total income, maternal employment, and education. These variables and their measured categories are also presented in Table 1. Maternal employment was based on the mother's report of her work situation during pregnancy or shortly after giving birth. Demographic factors included the mother's age at the birth of the child, race, Hispanic ethnicity, and if the mother was born outside the United States. Region and metropolitan area were also measured. In addition, religion and marital status at the child's birth were included in the model (Table 1).
Maternal health consciousness measured whether or not the mother smoked during her pregnancy. In addition, 2 other attitudinal indicators were included—whether or not the birth was reported as wanted by the mother, and whether or not she reported that she wanted any more children. Birth characteristics included whether the birth was the firstborn, if the delivery was a C-section, and if the birth weight was below 5½ pounds (Table 1).
Because the dependent variable was dichotomous (measured as 1 if breastfed, 0 otherwise), the breastfeeding model was estimated using logistic regression techniques. Logistic regression was appropriate in this case because the logistic curve remains within the 0 to 1 range, as opposed to linear regression, which produces predicted values <0 and >1. In our analyses, the coefficients represent the increase or decrease in the log odds of ever breastfeeding (versus not) associated with a unit or category change in an independent variable. Taking the exponent of the coefficients (log odds) gives the odds ratios (odds of breastfeeding versus not) and it is these odds that are reported in the tables.
Sampling in the NSFG involves a complex sampling design, therefore, it was necessary to consider sample weights in the estimation of standard errors. Such weights adjust for the probability of sample selection. Analyses were performed both with and without the weight adjustments on the standard errors (available from the authors) and no appreciable differences were found. Thus, significant effects reported in tables do not seem to be the result of nonrandom sampling.
After the presentation of descriptive statistics, we first examined the individual effect of each independent variable on the likelihood of ever breastfeeding. Next we estimated the combined effect of all the variables on the likelihood of breastfeeding in a multivariate model. We then eliminated the nonsignificant variables to determine the best predictive model with the fewest variables. Finally, we examined the relationship between breastfeeding and racial differences in infant mortality.
Descriptive statistics presented in Table 1 indicate that about 57% of the mothers breastfed their infants—percentages are higher for white mothers (65%) compared with black mothers (30%). If all single-live births reported in 1995 in the NSFG (N = 14 596) are included (available from the authors), the percentages are somewhat lower—48% of all births were breastfed, with 56% of white infants and only 23% of black infants breastfed. Because breastfeeding rates have been increasing over time in the United States,5 our sample rates based on recent breastfeeding experiences (births within 18 months of the survey date) are somewhat higher than rates based on all births reported by women of childbearing age (which group includes births as far back as 2 decades before the survey date). Again, based on all births, of those who were breastfed, the median duration of breastfeeding was 5.25 months for white infants and 3.38 months for blacks. Thus, overall, breastfeeding levels were lower among black women compared with nonblack.
Racial differences in socioeconomic background are also noted in Table 1. Over half of the black women in the sample had total household incomes below $16 000, compared with only 23% of the white mothers. More black mothers were not employed and had less than a high school education compared with white mothers. More than half of the black mothers were also single and never married, compared with only 11% of the white mothers.
White mothers were more likely to reside in the West compared with black mothers, and black women were more likely to live in central cities than their white counterparts. Black mothers were also more likely to report that they did not want any more children (60%) compared with white mothers (43%), and to report that their recent birth was unwanted (14% of black women compared with 7% of white women). Black infants were also more likely to be low weight at birth (11%) than were white infants (4%). Thus, based on the literature review, white mothers were more likely to have characteristics associated with the decision to breastfeed—higher income and education, marriage, living in the West, and wanting their birth—than were their black counterparts.
To examine the effect of these various characteristics on the likelihood of breastfeeding, we first estimated the individual effect of each maternal and birth characteristic on ever breastfeeding as reported in Table 2. With the exception of parity and the type of delivery, each of the characteristics significantly influenced the likelihood of breastfeeding (see column 1,Table 2).
Column 2 in Table 2 presents the multivariate model or the combined effect of all variables on the likelihood of breastfeeding. Many of the factors lost significance (as the effects neared zero) in the multivariate model, suggesting that the influence of these factors on breastfeeding is indirect. Additional analyses (available from the authors) suggest that maternal employment and marital status influenced the decision to breastfeed through household income (working mothers had higher household incomes than nonworking mothers, as did married women compared with single women).
The nonsignificant factors were removed from the model to produce the best fit with the fewest variables, and this final model is presented in column 3. Differences between the −2 log likelihoods reported for the multivariate model (column 2) and the simplified model (column 3) were not significant—indicating that the simple model fits the data as well as the full model. Thus, based on the odds reported in column 3, the likelihood of breastfeeding increased with income. In addition, women with a college education were almost twice as likely to have breastfed than were women with only a high school degree or less. Foreign-born women were about 75% more likely to have breastfed than US-born women, and women living in the West were >3½ times more likely to have breastfed compared with women living in other regions of the United States. Women were also more likely to not breastfeed if they considered the infant to be their last child; mothers not wanting any more children were over 1½ times less likely to breastfeed their child than were mothers wanting more children.
Finally, even after controlling for background characteristics such as income, education, marital status, low birth weight, and residency, race continued to have an independent effect on the decision to breastfeed. Black women were 2½ times less likely to breastfeed than were white women (reduced odds –1/0.4 = 2.5). To examine the race effect further, we interacted race with the variables in the simplified model (available from the authors); however, none of the interactions were significant, suggesting that the factors influencing the decision to breastfeed reported in column 3 do not vary by race.
Women who did not breastfeed were given options in the NSFG to indicate why. Although restricted to the categories provided in the survey, these responses give some indication of women's reasons for not breastfeeding. Table 3 provides the responses of the women sampled who did not breastfeed their child (43% of women with a child 18 months and younger in 1995). Only 10 percent of these mothers indicated that they did not breastfeed because of employment conflicts, and another 14% reported having medical or physical problems that kept them from breastfeeding. The majority indicated that they chose not to breastfeed because they “preferred bottle-feeding” as opposed to breastfeeding. Racial differences in preference were also noted—83% of black mothers not breastfeeding reported a “preference for bottle-feeding” compared with 62% of white mothers. This suggests that black women bottle-feed their infants not because of employment or physical difficulties, but because of preference.
Breastfeeding and Infant Mortality
To further examine racial differences in breastfeeding, we combined the birth interval files from the NSFG Cycle IV, 1988, and the NSFG Cycle V, 1995 data to relate racial differences in breastfeeding to infant survival. This analysis included all single births that lived with the mother reported in 1988 and 1995 (N = 24 566). Using all single, live-births, we estimated the effect of race on infant mortality before age 1 year. We then included ever breastfed and low birth weight in the models to see if they accounted for the racial difference in mortality.
Past studies have shown that racial differences in infant mortality in the United States are attributable, in part, to the high incidence of low birth weight infants in the black population.26,,27 If low birth weight explained the racial difference in infant mortality, then the race effect would disappear once low birth weight was included in the model. The same would be true for breastfeeding if it accounted for the racial variation in infant morality. Because the infant mortality literature examines primarily low birth weight and not breastfeeding as an explanation for differences in black-white infant mortality, we wanted to explore this issue given the lack of breastfeeding by black women.
Our models of infant mortality were estimated using survival tables and Cox regression techniques; Table 4presents the effect of the independent variables on the odds of death before age 1 year. Infants under 1 year of age at the date of interview were included in the analyses up to the number of months they had lived by the interview date. In addition, infants living for less than a month were not included in the analyses because, given the limitations of the data, it was not possible to determine if infants died shortly after birth (that is, before breastfeeding could be initiated) or later when breastfeeding could have begun. Given this data limitation, we could not identify deaths that could have been preceded by breastfeeding from those that could not. It is assumed that deaths after 1 month of age occurred after breastfeeding could have been initiated. Including deaths before 1 month of age would have overestimated the effect of breastfeeding on infant mortality because deaths occurring hours after birth before breastfeeding could be initiated would be attributed to not breastfeeding and its effect on infant mortality.28 By excluding these deaths, our estimates of the effect of breastfeeding on infant mortality were more conservative.
Census Bureau Statistics29 for 1995 indicate that the infant mortality rate for blacks was 15.2 deaths before age 1 per 1000 live-births compared with 6.3 for whites. As shown in Table 4, model 1—black infants were almost 1.5 times more likely to die before the age of 1 year than were nonblack infants. According to past research, low birth weight is a primary factor influencing infant mortality rates, with black births more likely to be low weight than white births.26,,30 Therefore, low birth weight was introduced into the model in addition to race. As shown in Model 2, after controlling for low birth weight, the race effect is reduced and becomes nonsignificant. Model 3 then introduces ever breastfed into the model with race and after controlling for breastfeeding, the effect of race is reduced even more; thus, breastfeeding accounts for as much of the race difference in infant mortality rates as does low birth weight. Model 4 shows the additive effect of low birth weight, breastfeeding, and race on the likelihood of death before age 1. Low birth weight infants are 4.3 times more likely to die before age 1 than are normal weight infants, and infants that are breastfed are 80% less likely to die before age 1 than are never breastfed infants.
We found a strong race difference in the decision to breastfeed, even after controlling for socioeconomic background and birth characteristics; black women were only 40% as likely to breastfeed as nonblack women. This finding concurs with past research4,,19 and although breastfeeding rates have begun to increase among those populations traditionally least likely to breastfeed,5,,10 the majority of black women still prefer bottle-feeding over breastfeeding.
The primary response given by black women in our study for not breastfeeding was a “preference for bottle-feeding”; 83% of black women not breastfeeding chose “preferred to bottle-feed” as the response category explaining why they did not initiate breastfeeding. Our understanding of this “preference for bottle-feeding,” however, is limited given the structure of the survey response options in the study. Additional study is needed to determine the psychological and social reasons behind preferences for bottle-feeding. Our findings, however, do suggest that breastfeeding explains the racial difference in infant mortality as well as does low birth weight. Thus, by increasing breastfeeding among black women, the racial gap in infant mortality should narrow—a gap that is currently (1997) about 1.3 times higher for blacks than whites.29
Our modeling of the decision to breastfeed indicates that, in addition to the race effect, college-educated women and women living in the western part of the United States are the most likely to choose breastfeeding—characteristics not very representative of the black population in the United States. Past studies note that more highly educated women recognize the benefits of breastfeeding and are more likely to choose breastfeeding as opposed to bottle-feeding.6–12 In particular, Wright6concluded that better educated women are more likely to breastfeed, to breastfeed exclusively, and to delay the introduction of formula compared with less educated women. In addition, previous research has found that higher education increases the likelihood that women will breastfeed regardless of race.19 Thus, more efforts are needed to target black women generally, and poor black women in particular. Past studies note the importance of health care providers in influencing the decision of women to breastfeed.9 If physicians and other health care workers can successively increase the percentage of black infants being breastfed, the infant mortality rate between blacks and whites should close.
In addition, our findings indicate that women are less likely to breastfeed their last child if they state that they don't want any more children. Pediatricians, as well as other health care providers, could especially target this group and encourage women to breastfeed their infants when they report they have finished childbearing. Thus, particularly if they can target high-risk women, health care workers can play a successful role in promoting breastfeeding.9
Additional study is needed to clarify the strong regional effect in breastfeeding—why do women in the western states breastfeed more than do women in other regions? Whether this is attributable to differences in the medical community, more role models, or more community support is unclear. However, in particular, additional research is needed to better understand the factors that inhibit breastfeeding among black women and the factors that promote a preference for bottle-feeding. Based on our results, health care providers need to target the black community so that the “breast is best” message replaces a preference for bottle-feeding. Such efforts seem to be as critical in reducing black infant mortality as is targeting low birth weight infants. Thus, the encouragement of breastfeeding needs to be a priority for physicians and health care providers in furthering the goal to promote child health.
- Received July 17, 2000.
- Accepted November 14, 2000.
Reprint requests to (R.F.) Department of Sociology, Brigham Young University, 852 SWKT, Provo, UT 84602. E-mail:
- NSFG =
- National Survey of Family Growth
- Forsyth JS
- Kurinij N,
- Shiono PH,
- Rhoads GG
- ↵Ryan AS. The resurgence of breastfeeding in the United States.Pediatrics. 1997;99(4). URL:http://www.pediatrics.org/cgi/content/full/99/4/e12
- Wright A,
- Holberg C,
- Taussig LM,
- the Group Health Medical Associates Pediatricians
- Lindberg LD
- Ryan AS,
- Martinez GA
- Joffe A,
- Radius SM
- Simopoulos AP,
- Grave GD
- Ryan AS,
- Rush D,
- Krieger FW,
- Lewandowski GE
- ↵Martinez GA, Dodd DA. Milk feeding patterns in the first twelve months of life. Pediatrics. 1983;1981:17:166–170
- Samuels SE,
- Margen S,
- Schoen E
- ↵Murphy SL. Deaths: Final Data for 1998. In:National Vital Statistics Reports. Vol. 48. No.11. Hyattsville, MD: National Center for Health Statistics; 2000
- ↵US Census Bureau. Statistical Abstract of the United States: 1999. 119th ed. Washington, DC: Government Printing Office; 1999. No. 133
- Copyright © 2001 American Academy of Pediatrics