The survey also included questions regarding drinking patterns and questions to ascertain the demographics of the study participants.

Data Collection for the Quantitative Study
The second phase of data collection used the questionnaire developed from the focus group process. The purpose of the study was explained to those who volunteered to participate and the questionnaire was distributed to women agreeing to participate and who were attending the Black Infant Health and Women Infant and Children (WIC) classes on the days designated for data collection. Respondents were informed that consent to participate in the study was established by completing the survey. To encourage honesty in survey responses, respondents were informed that no identifying information was included in the survey, therefore, the survey responses could not be tied back to any particular respondent. In addition, respondents were instructed not to include their names on the questionnaire. As an incentive, women received a five-dollar gift certificate upon completion of the survey.

Data Analysis for the Quantitative Study
Responses to the questionnaire were entered into the Statistical Package for the Social Sciences (SPSS) for analysis. Surveys were excluded from the analysis if they were less than 80% complete and if the participant failed to meet inclusion criteria for alcohol use prior to or during pregnancy. Of the 179 collected a total of 148 met the inclusion criteria. Women who reported never having consumed alcohol were excluded from the analysis. Frequency statistics were done on all variables. Missing values were imputed with the expectation maximization algorithm for cases not missing more than 20% of responses. All available variables were used for imputation in accordance with Schafer and Graham’s (2002) recommendations. Due to low participant responses for normative beliefs only 5 of the 23 normative beliefs, identified via the focus groups, were included in the quantitative analysis.

Several questions were included in the survey to identify intention, attitude, subjective norm, and perceived control. Cronbach’s alpha were calculated for the multiple questions regarding intention, attitude, subjective norm, and consistency on the questions. A multiple regression was run for the theory of planned behavior variables to determine which elements of the theory predict the intention to quit. The theory of planned behavior is based on expectancy value models which require measurement of two variables for each salient belief underlying the attitude, subjective norm and perceived control and that products be formed from each of these pairs. For example, each outcome belief is multiplied times the value of that outcome. However, the theory also suggests that the beliefs and values can be scaled by adding or subtracting a constant prior to multiplying the two. For example, if the outcome value had been measured on a scale running from 1 to 7, 4 might be subtracted from the value chosen by each respondent to yield a value score running from -3 for a rating of a negative valued outcome to +3 for a positively valued outcome. Since different scaling procedures give different results when the product terms are formed, Ajzen (2006) recommends an optimal scaling procedure which we used to maximize the correlation between the product terms and the intended prediction. Some researchers (e.g., French and Hankins, 2003) have objected to this use of optimal scaling as being overly complicated and theoretically unclearly justified but we elected to follow Ajzen’s recommendations.

One problem with optimal scaling is that numerous scaling parameters can produce very similar correlations with the outcome variables and some of these may be positive and some negative. For example, in our study one outcome was that the mother would be healthier if she abstained from drinking. Participants rated the likelihood of that outcome and its value. The optimal scaling procedure resulted in scaling parameters of -27.499 for the outcome and -22.715 for its value. This produced a correlation of -.4204. However, if the scaling parameters were set to +30 and +18 the resulting correlation was +.4196—a correlation with a difference in absolute magnitude from the optimally scaled amount of only .0008 but completely different in direction. Thus, different scaling parameters can produce essentially the same magnitude of correlation but different signs for the correlation. How should we choose the direction of the correlation? We chose the sign based on the simple correlations of the outcome likelihood by itself and the outcome value by itself with the attitude. In the example just discussed, both of these had positive correlations so we chose a positive sign. This process also had the advantage of making conceptual sense as one would expect that mothers who believed that abstention would make them more healthy would have a positive attitude toward abstention rather than the negative attitude that a strict adherence to the optimal scaling procedure would suggest. A similar procedure was followed for subjective norms and perceived control.