This study attempted to use the TPB to explain variance in breastfeeding intentions and behaviors among a cohort of Midwest breastfeeding mothers. To do so, the research team used SEM to fit constructs within the TPB as latent variables and model breastfeeding behaviors. The constructs, however, did not hold up when modeled in this population. These issues with poor model fit may have gone unnoticed had the research team not opted to use SEM. This is especially apparent when considering construct alpha scores. The alpha scores for each construct with the TPB would have been acceptable, or nearly acceptable, for a traditional regression model (α > than 0.7). However, latent variables can parse measurement error more accurately. Alpha scores are based upon correlations among variables, but McDonald’s [19] omega score is based upon item factor loadings. In this way, unlike alpha, omega scores do not assume that all items contribute equally to constructs or that item errors are not correlated with one another [24]. When all observed variables measure the construct in the same way, as assumed in an alpha score, the omega score would be identical to an alpha [21]. In the case of the present data, the omega scores for the constructs are unacceptably lower than the alpha scores, indicating that they are less reliable than would otherwise be known (see Table 2).
Unlike Duckett and colleagues [4] and Dodgson and colleagues [25], the TPB did not fit these data. This difference may not mean the TPB is an entirely invalid model, but it does indicate that there are problems that question the reliability in its present form. The authors’ primary reasoning for this difference could be related to current SEM methods since statistical packages and computing power have advanced. Additionally, Dodgson and colleagues [25] appeared first to create summed scales and then produced their SEM with the resulting correlation matrix. The problem, however, arises before the point of creating additive scales. This method masks the fact that the items in the scale do not belong together. This finding can go unnoticed as alpha scores can border on acceptable, but using an omega score that does not rely on the same assumptions flags problems. Similar to the findings in this study, Dodgson and colleagues’ alpha score on control was their lowest, although in their case alpha scores were higher and within more acceptable ranges. In our case, this is notable because we were also within an acceptable alpha range (α=0.68) but the omega flags problems with a score of only 0.34, which is very low. In other words, the problems highlighted here have more to do with the items theorized to belong together within each domain. Yet, the authors remain agnostic regarding the possible overall theorized relationships as they may remain effective.
Guo and colleague’s [7] meta-analysis is a welcome addition to breastfeeding behavior research as understood through the TPB. However, the results rely heavily on published correlation matrices that make use of additive scales that implicitly and statistically assume equal item variance within construct domains. The selected methods inadvertently obfuscate how poorly the unreliable instrument fits the lived experience, confirming the need to refine and update the instrument.
For some time, research focused on the formation of beliefs and values and the actions that flow from socialization has highlighted the importance of close ties. Exposure to relationships, social structures, and one’s position within them, shape perceptions of the world that are then replicated in behaviors [26, 27]. Typical Midwest family structures and relational ties within the Midwest, and this location, specifically, likely have a large influence on mothers’ perceptions of default desirable behaviors, including breastfeeding behaviors.
Study participants were primarily recruited via social media (Facebook), indicating this cohort of childbearing women is active on social media and, at least in this case, social media activity is connected to actions. Recent research has shown that women are more active in seeking out health information than men and that the internet plays a key role in this information consumption [28]. This idea holds true to seeking specific information tied to breastfeeding [29,30,31,32]. As such, the content and slant of breastfeeding information presented on social media is likely a key unmeasured variable for the patterning of breastfeeding behaviors. However, the questionnaires utilized in this study failed to gather data related to the influence of social media and social networking sites on the attitudes, subjective norms/normative beliefs, perceived behavioral control, and intentions toward breastfeeding behaviors.
While the overall efforts to fit a SEM for the TPB failed, variables related to formula feeding clustered together quite well in this study. Beyond the fact that formula measures factoring well is conceptually notable, this may also be a small indication that the failure of TPB to fit this data well is not merely a data artifact. After all, it would not make sense to say it is merely a data problem when some manifest measures that clearly belong together (formula feeding questions) statistically do hold together quite well. If there were a fundamental problem with the data, it would be odd for it to not be widespread throughout the measures. The uniformity of messaging by the formula industry may explain why those items may more clearly factor out together. Strong, uniform messaging is also a key reason why mothers may discontinue breastfeeding [33]. Although formula marketing is down in the US, it may be that there remains a long-standing impact on formula feeding attitudes. In contrast, breastfeeding messaging remains uncoordinated and unclear, even among health care providers [34, 35].
Limitations
The limitations of this study include convenience sampling, a limited sample size, homogeneity of the study sample, and the use of previously developed questionnaires with a limited ability to capture the constructs under investigation. Despite these limitations and null findings, this study remains particularly valuable to nursing science, in which the development of interventions is driven by theory. Among dissertations focused on breastfeeding research during the last 10 years, approximately eight dissertations utilized the TPB as a guiding theory. This does not encompass the multitude of studies in maternal and child health currently underway, nor published manuscripts using the TPB as a guiding framework.