Translation and validation of the Breastfeeding Motivation Scale in China

Mothers’ sociodemographic and clinical data were collected by investigators using the questionnaire developed by the researchers. The mothers’ sociodemographic and clinical data included (1) sociodemographic data, (2) gestational age, (3) pregnancy and birth history, (4) mode of birth, (5) medical history, and (6) neonatal situation.

The BMS was developed by Israeli academic Miri Kestler-Peleg and his team members [20] in 2015. The original English-language BMS initially contained 24 items, but after principal axis factor analysis, the developers removed item 1, "Breastfeeding is more convenient because there is no need to wash and sterilize bottles, and breastfeeding can be done anywhere, anytime". Therefore, the final BMS in English consists of 23 items covering 5 subscales: (1) enjoyment and bonding, (2) maternal self-perception, (3) significant others' pressure, (4) baby's health, and (5) instrumental needs. Each item is scored on a 4-point Likert scale ranging from 1 meaning “strongly disagree” to 4 meaning “strongly agree”. Respondents are asked to rate extent to which each item matched the reason they chose to breastfeed. The scale has no total score. As the subscale score increases, the motivation that represents that subscale also increases [23]. The Cronbach's α for each subscale of the BMS in English ranged from 0.62 ~ 0.93. The Cohen’s kappa consistency coefficients of the scale ranged from 87.5% to 100 percent.

The final English version of the BMS was clarified with the author via email. The translation of the BMS closely followed the guidelines of the principles of good practices for translation and cultural adaptation of measures established by the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) [26]. Phase 1 involved four steps. Step (1) Translation: Two bilingual native Chinese speakers who were also fluent in the English language were asked to independently translate the English version of the BMS and the accompanying instructions provided with the scale from the English to Chinese language. The two translators were a Ph.D. in nursing who worked in the US and a Ph.D. in obstetrics who had returned from studying in the US. The two translations were then assessed by two native Chinese speakers online, who reached a consensus on any discrepancies to produce a single translated scale in Chinese named Chinese version 1 of the BMS. Step (2) Back-translation: Two translators with knowledge of both the Chinese language and culture independently back-translated the Chinese version 1 of the BMS to English. The two translators were a Ph.D. in obstetrics and gynecology nursing who studied and worked in the United States and an English teacher from a popular university in the U.S. The two translators discussed the discrepancies of the back-translation and reached a consensus. Then, they synthesized the details of the back-translation to produce a single back-translated scale in English, which was named the back-translation version 1 of the BMS. The researchers sent the Chinese version 1 of the BMS and the back-translated version 1 of the BMS to the original author of the BMS to ask for his advice. The original author proposed changes in three items of the scale, and two modifications (“among us” represents the closest circle in item 3 and only mothers can breastfeed the baby in item 10) were made based on the author’s advice. Then, we reached a consensus to produce Chinese version 2 of the BMS. Step (3) For cross-cultural adaptation, five experts who majored in obstetrics and were skilled at translation (a nursing specialist familiar with cross-cultural adaptation, a nursing management specialist, an obstetric nursing specialist, an obstetrical clinician and an international board-certified lactation consultant) were invited to evaluate the accuracy of the translation, whether the translated version was clear and easy to understand, and whether it was consistent with the cultural background of Chinese people. The experts rated each item on a 4-point Likert scale ("very relevant" = 4, "relevant" = 3, "not very relevant" = 2, "not relevant at all" = 1). They provided few comments, and no further modifications were needed. Then, Chinese version 3 of the BMS was produced. Step (4) To perform an initial evaluation and to assess the understanding of the Chinese version 3 of the BMS in the Chinese population, a small sample pilot study was conducted. Twenty-five native-Chinese-speaking mothers were recruited after labor at the hospital according to the inclusion and exclusion criteria to fill out a sample scale (Chinese version 3 of the BMS). Cognitive interviews were also conducted to collect modification suggestions from respondents. Cognitive interviews are a common method used to pretest whether survey questions are understandable and answerable. The 25 mothers were encouraged to describe their thoughts while answering the scale questions, which could be done through think aloud, verbal probing, and paraphrasing methods, etc. [27]. We developed a semi structured interview guide to collect further information about the scale from the women. The interview questions were as follows: Can you repeat this question in your own words? What does that word mean to you? Tell me more about your thinking when you think about the question. Are there questions you believe should be modified? Why? What do you think the word in ‘...’ could be adjusted to make the question more understandable? Are there questions you believe should be deleted? Why? Any other questions? The interviews were recorded via audio, were held in the meeting room of the ward and were conducted by an investigator who was trained in cognitive interviewing. Then, the translation team modified Chinese version 3 of the BMS according to the interview content, which the investigator summarized after the interviews. Generally, respondents reported that the sample scale could not be misunderstood, and four slight changes of items were made after the interviews (see Table 1 for the modifications made after the interviews). Finally, Chinese version 4 of the BMS was formed. This was the final Chinese version of the BMS before psychometric property testing. We call Chinese version 4 of BMS ‘the Chinese version of the BMS’ below.

After obtaining permission to carry out the study from the human research ethics committee of Shanghai First Maternity and Infant Hospital, we oriented the investigators and registered nurses working in Shanghai First Maternity and Infant Hospital to clarify the details of the survey, including the team members, survey tools, time, communication method, data collection methods, and so on. The questionnaires were administered to the mothers from 1 December 2021 to 1 July 2022. To avoid affecting the clinical work of the hospital and the nonparticipating mothers, we collected the questionnaires from 3:00 to 5:00 p.m. to reduce the possibility of survey interruption and to reduce the number of invalid questionnaires. Before the survey began, the investigators explained the study protocol to the mothers again. The mothers were instructed on how to understand the questionnaires and how to complete them.

The questionnaires were collected and checked by the investigators on site, i.e., in the maternity wards. The mothers completed an online version of the self-administered questionnaires. The mothers were asked to scan a QR code to fill out the questionnaire, and all responses were anonymous. It took approximately 15 min to complete the questionnaire, and this amount of time was assessed as adequate. Twenty mothers were randomly selected to answer the Chinese version of the BMS twice, with second time being 2 weeks after the first to assess the test–retest reliability during the postpartum 2-week visit. This 2-week duration was proposed by Stewart and Bass because it is difficult to affect a person's memory and practice in 2 weeks [28]. The participating mothers’ telephone numbers and gestational age were recorded when the data were collected for the first time. We invited the five experts for the two rounds of expert consultation to test the content validity as mentioned above. All experts had bachelor's degrees or above.

The internal consistency coefficient [28] of the Chinese version of the BMS measured by Cronbach's α was 0.887. The Cronbach's α values of the subscales of the Chinese version of the BMS were 0.876, 0.821, 0.857, 0.900, and 0.849, respectively. These values revealed that the Chinese version of the BMS was acceptable for all subscales.

The validity of the Chinese version of the BMS was tested using expert consultation and confirmatory factor analysis.

After two rounds of expert consultation, the content validity of the scale was calculated. The item-level content validity index (I-CVI) of each item in the Chinese version of the BMS ranged from 0.8 to 1.0, while the scale-level content validity index (S-CVI) was 0.83, with a scale-level content validity index average (S-CVI/Ave) of 0.97.

The hypothesized measurement model of the Chinese version of the BMS comprised 23 items across 5 factors: enjoyment and bonding (8 items), maternal self-perception (6 items), significant others’ pressure (4 items), baby’s health (2 items) and instrumental needs (3 items). The Kaiser‒Meyer‒Olkin value was 0.952, with P < 0.001, and the significance of Bartlett's sphericity was 0.000 (χ² = 3680, df = 253). These results supported proceeding with the factor analysis.

The construct validity, convergent validity and discriminant validity were estimated using the maximum likelihood estimation method [30]. A 5-dimensional model identical to the structure of the original version of the BMS was proposed with the aim of checking whether the model was adequate. The goodness of fit statistics was χ²/df = 2.4, P < 0.001, RMSEA = 0.08, CFI = 0.91, IFI = 0.92, and TLI = 0.90, indicating an acceptable model fit. The standardized factor loadings were all greater than 0.60 (P ≤ 0.001) (Fig. 2). All the items loaded significantly onto their respective factors. No variables were deleted from the original model in this study.

As a type of construct validity, convergent validity is related to whether a latent variable is well estimated with the selected indicators (i.e., tested variables should have high correlations with supposed similar constructs) [34]. Statistical evidence of convergent validity was checked through the convergent validity coefficient (i.e., composite reliability) and average variance extracted (AVE) [35, 36]. The formula \(\frac{{(\sum Factor\;loading\;value)}^2}{{(\sum\;Factor\;loading\;value)}^2+\sum\;measurement\;Error}\) was used to calculate the composite reliability (CR), and the formula \(\frac{{\sum\;Factor\;loading\;value}^2}{{\sum\;Factor\;loading\;value}^2\;+\sum\;measurement\;Error}\) was used to calculate the AVE [34].

The CR has to be 0.70 or greater and the AVE has to be 0.36 or greater for acceptable convergent validity [37, 38]. As shown in Table 3, the CR values corresponding to each latent variable were 0.733 ~ 0.926, and the AVE values were 0.476 ~ 0.653.

Discriminant validity is also a type of construct validity. It demonstrates that different or unique constructs do not correlate with each other or have low correlations with tests from which it should differ [36, 39]. The discriminant validity was used to determine whether the five indicators of ‘enjoyment and bonding’, ‘maternal self-perception’, ‘significant others’ pressure’, ‘baby’s health’ and ‘instrumental needs’ were distinct factors from one another for the current study. Generally, discriminant validity was assessed through confirmatory factor analytic models with every pair of latent constructs using AMOS [40]. To verify the discriminant validity of the Chinese version of the BMS, the correlations among the five measured variables needed to be lower than Pearson r = 0.85 [34], and the square root of average variance extracted (AVE) from the variable needed to be greater than the interconstruct correlations between the variable and other variables in the model [35]. Table 4 provides strong evidence of discriminant validity, with correlations among variables (P < 0.001) (off-diagonal elements) and the square root of AVE on the diagonal.