We developed a pool of items on FP based on an extensive review of published articles in related topics in general and in the pastoralist community in particular. The tool consisted of three parts; (1) 13 knowledge items (2) 12 items of the perceived male involvement on FP and 73 items on perception (items of an integrated behavioral model (IBM)). The IBM items contain expressional and instrumental attitude, subjective norm, intention to use, self-efficacy and perceived control. Tool development passes a lot of steps. To mention its process, in the beginning, experts in reproductive health and Health education assured the face and content validity of the tool using two-panel discussions. Items that were not relevant were dropped while those reported as not clear were re-worded. On the next step, the tool was piloted in 10% (90 women) of the sample. Items with negatively worded were reversed scored. And reliability test also was done and items with low-reliability value (Cronbach alpha <0.7) were excluded from the list. Exploratory factor analysis was conducted to select relevant constructs. For the IBM tool, the total score is the sum of responses summed to get the total score. Items with high scores indicate a more positive relation towards FP use. An extensive revision of the tool was made by removing items with a low score and adding other items. The revised tool was pretested in 5% (45) of the sample. In the end, reliability test, exploratory and confirmatory factor analysis was done on 891 tools. The data was collected in two phases as a baseline and end-line data. Also, the construct validity (concurrent and discriminant validity) test was checked using EFA.
Study participants, Sample size and sampling procedures
The study employs pastoralist married women of reproductive age group (15-49) years who reside in the Afar region. Married women who were critically ill during the data collection were excluded from the study. A cluster sampling technique was used to approach the study participants. Three districts namely: Kori, Afambo and Mille were included in the study. From each district, 297 married women were approached. The total sample size was proportionally allocated to the selected kebeles (11 from each district). In the end, a systematic sampling technique was used to approach the study participants in the household. Based on the sample fraction, women were selected at equal interval using systematic random sampling
Data collection procedure
The tools contain three parts 1) knowledge on FP, 2) perceived male involvement on FP use, 3) IBM related items: intention to use, subjective norms, self-efficacy, perceived control, expressional and instrumental attitude. The tool was piloted in 10% of the sample after it was developed by reviewing different literature. The collected piloted tool was exposed to a reliability test. Then, the reliable tool was done for exploratory and confirmatory factor analysis. After all necessary modifications followed the piloted test, the tool was pretested in 5% of the sample. Modifications were made based on the pretest finding. As the tool was initially developed in English; it was translated to the local language (“Amharic”) and back-translated to English to ensure consistency. Two-person who were blind to each other was used in the translation and back-translation process. The expert on reproductive health, health behavior and promotion including the researcher were analyzed the translation of each item critically and checked the compatibility of the translated items. The tool composed of positively and negatively worded items to indicate the respondent’s agreement or disagreement. Items in one category example, knowledge towards family planning use had the same scored. Pastoral married women were approached in the household settings and oral consent to participate in the study was sought. Before interviewing the married women, the data collectors explained the purpose of the study, sampling procedure, ways of data collection (open data kit (ODK)), right not to respond to all or segment of the questionnaire and assured the participant's response’s and the response would be kept strictly confidential, no one except the research member would have access. To collect the data six clinical nurses and two supervisors with MPH degrees were used. The training was given for data collectors and supervisors. Data were collected using an open data kit (ODK) a mobile-based application.
Intention to use of FP is defined as the motivational factors that influence a given behavior where the stronger the intention to perform the behavior, the more likely the behavior will be performed. A total of 8 items with response scored 1(uncertain /Disagree) to 3(Certain/Agree). All the responses of intention to use FP were added together to produce one combined score.Besides, perceived male involvement in FP was collected using 12 items with a response category ranging 1(Disagree) to 3(Agree). Hence, its responses were added together to produce one combined score.
The items or constructs of integrated behavioral models (IBM): expressional and instrumental attitude, subjective norm, self-efficacy and perceived control were developed and conceptualized to the pastoralist setting. All the components of IBM had two components: direct and indirect measurement. Expressional Attitude (EA) or affect is the married woman's emotional response to the idea of performing the behavior(FP use). Eight items with response categories ranging from 1(unlikely/unworthy) to 3(likely/worthy) were used. Instrumental attitude (IA) or cognitive is the married women's evaluation for the FP use . It was determined by belief about the outcomes of behavior(FP use). Sixteen items response categories ranging from 1(uncertain /unlikely) to 3(Certain/likely) was used. Subjective norm (SN) is the belief about whether most people approve or disapprove of the behavior(FP use) in their community. With a response scored 1(uncertain /unlikely) to 3(Certain/likely) was used to collect twenty-two items. Perceived control (PC) is a married woman's perceived control over behavioral performance(FP use). Ten items were scored from 3(highly/agree) to 1(doesn’t matter/disagree) was used. Self-efficacy (SE) is the personal belief that married women can successfully perform a specific action (FP use) under specified conditions. It was measured with 7 items. Responses were coded as 1 (disagree) and 3 (Agree) (9, 10).
Furthermore, the indirect measurement (IM) for the items of subjective norm, expressional and instrumental attitude had a belief and evaluation components. Multiplying the belief response with its corresponding evaluation was made to create a continuous variable. And, the multiplied items of the response were summed up. Along with this, the response of self-efficacy and perceived control was added to form a continuous variable. It should be noted that the IBM constructs had a direct measurement (DM) with a value ranging from -2(“poor/low”) to 2(“good/high”). Hence, a total of 12 (EA), 4 (IA), 17 (SN), 5 (PC) and 4(SE) tool was used. Finally, a correlation test of the DM with its corresponding IM was calculated to assess the importance of the IM to measure the IBM constructs related to FP use in the pastoralist setting.
Data Quality Control
To assure the quality of the collected data the following points 1) training for data collectors and supervisors 2) strong supervision at the field 3) an electronic mobile-based application (ODK) was used. Along with, in the pastoralist area where the majority of the married women unable to read and write a simple sentence they may face difficulty in realizing the difference between one item to another. However, a maximum effort was done by the data collectors to encourage her to ask for any ambiguity in the items and to clarify the item based on her view. Hence, additional clarification on the items was done in the case of any deviation from the right content of the tool.
To analyze the data AMOS of SPSS 22 for windows (SPSS Inc. Version 20., Chicago, Illinois) and R software version 3.6.1 was used. Accordingly, the data were analyzed for 1) face and content validity,2) reliability test 3) factor analysis 4) independent sample t-test and correlation test.
Face and content validity
The face and content validity were done based on expert opinions of reproductive health (4) and health behavior and promotion (1) specialists who have experience in family planning research and pastoralist community. In the beginning, an effort was made to experts to have clear expectations and understanding about the tool development. Apanel discussion was made with the experts to forward their constructive comments for the enrichment of the drafted tool. Hence, to facilitate smooth communication and a high response rate in the tool development, a face to face approach through an expert panel meeting was organized. At the first step (face validity) of the developed tool was given to the experts to look at the items and agreed the tool was valid to measure FP in the pastoralist context. The second step (content validity) of the tool was assured by answering the question of whether the developed tool of FP fully measures or assess the FP issue in the pastoralist context. And, the experts are requested to critically review the domain of the tool. The comments and suggestions of the experts regarding the developed tool were documented and extensive revision was made. As a result, necessary modification to ensure the developed tool readability, clarity, and comprehensiveness was done. In the end, the revised version of the tool was distributed to the expert to add their additional suggestion and to reach an agreement in the developed tool. As a consequence of the discussion few points were raised. Hence, a revised version of the tool was developed after a necessary modification and corrective action was made.
It was determined using the internal consistency test. The internal consistency was calculated using Cronbach’s Coefficient Alpha. A Cronbach’s alpha higher than 0.7 was considered as reliable items . All items of the tool were separately subjected to reliability test and items were dropped till the Cronbach alpha coefficient was found to be greater than 0.7 to indicate the presence of acceptable consistency of items.
3) Factor analysis
In the whole process of our tool development, we employ exploratory and confirmatory factor analysis. It was cognizant that the exploratory factor analysis (EFA) was used in the earlier process of tool development, whereas confirmatory factor analysis (CFA) was used in the later phase of tool development to assess the factorial structure found in EFA and after the underlying structure has been established. Also, used for instrument development and validation to explain the result with good quality (11).
3.1) Exploratory factor analysis (EFA)
We used EFA to differentiate the unique and common variance and as a precursor to the confirmatory factor analysis(CFA) in scale development. For each item of the IBM constructs as well as the other indicators, EFA was done. Since it has no a priori restriction it was used to identify the pattern of relationship between the IBM constructs and other indicators to the latent variable. To identify the relevant indicators to be included in our tool adequate sample size (891) was included in our final model. Before factor analysis, the Kaiser-Meyer-Oklin (KMO) measure of sampling adequacy and Bartlett test of sphericity were calculated to check the suitability of the data for the factor analysis(12). The Kaiser-Meyer-Olkin (KMO) index was checked to assess the adequacy of the sample size, which was > 0.84, indicating an adequate sample size. KMO measure of sampling adequacy measure varies between 0-1 and values closer to 1 are better, if it is greater than0.5 then one can proceed for EFA. Bartlett’s Sphericity test along with its chi-square was significant (P<0.001), which verify the existence of sufficient correlation (coefficient greater than 0.4) among the items confirming of factorability of the correlation matrix. Along with this, the correlation matrix was assessed for multicollinearity (coefficient greater than 0.9) and singularity (coefficient=1) of items. Subsequently, multiple approaches were used in synergy to assist the decision on the number of relevant factors. Firstly, to the factor selection or determining the number of factors to be retained in the tool eigenvalue-based procedures; Kaiser-Guttman rule, parallel analysis, and scree plot were used. Factors with an eigenvalue of 1.0 and above were retained for further investigation. According to the Kaiser-Guttman rule, the eigenvalue less than 1.0, describes the variance explained by a factor that is less than the variance of a single indicator. Secondly, each of the eigenvalues of the factors was plotted and inspected to find a point at which the shape of the curve changes direction and becomes horizontal, with the point of shift indicating the number of factors. Thirdly, only the factors that explained a total cumulative variance of 60% and above were retained. As a factor extraction method, the principal axis factor (PAF) was used as being free of distributional assumption and being less prone to an improper solution than the maximum likelihood (ML) method (11). Apart from this, direct oblimin rotation with a delta value of zero was used with the assumption of factors are allowed to intercorrelate and helps to avoid misleading solution (12). Also, it is a is preferred rotation mechanism because it provides a more realistic representation of how factors are interrelated and it produces the same solution as an orthogonal solution in a situation where the factors uncorrelated. Items with a factor loading less than 0.4 were eliminated from the list of tools whereas an item with a factor loading above 0.4 was selected for the final selection of the scales. And items cross-loading above 0.40 were deleted (13). The appropriate name was given by the researcher for the retained items following their factors. Along with doing EFA, a construct validity (discriminant and convergent) was checked. An EFA results of factor loading of at least 0.40, no cross-loading of items above 0.40 was revealed discriminant validity, whereas an eigenvalue of 1 and a loading of at least 0.40 will satisfy the criteria of convergent validity(14).
3.2) Confirmatory factor analysis (CFA)
Confirmatory factor analysis (CFA) used in the later phases of scale/tool development to explain the result with good quality and to assess the factor structure found in EFA. Besides, t requires a strong empirical foundation to guide the specification of the model. In the CFA, factor rotation was not specified as it’s the nature of fixing cross-loading factors to zero. The result of CFA was summarized using indices: Root Mean Square Error of Approximation (RMSEA), Standardized Root Mean Square Error of Approximation (SRMSEA), Comparative Fit Index (CFI) and Tucker- Lewis Index (TLI). In addition to the above indices, a Normed chi-square (x2/ df) also used. The cut value for the indices differed across different literature. Accordingly, we use the following cut-value to determine the goodness fit of our model. With this in mind, RMSEA lower than 0.08, SRMSEA less than 0.08, CFI greater than or equal to 0.90, TLI greater than or equal to 0.95 and chi-square(normed chi-square) with lower value was considered as a significance test and goodness of fit in the CFA (13, 15).
4) Independent t.test and correlation test
Independent sample t-test was used to find the difference in mean of perceived male involvement, subjective norm, self-efficacy, perceived norm, expressional and instrumental attitudes between FP users and non-users. In addition to the independent test, Cohen’s d was calculated. Cohen's d is an appropriate effect size for the comparison between two means. A d of 1 indicates the two groups differ by 1 standard deviation (SD) and with 2 d’s value indicates a 2 SD difference. Its value ranges from 0 to infinity: d=0.2 (a 'small'), d=0.5( 'medium' ) and d=0.8 (a 'large') effect size. This means that if two groups' means don't differ by 0.2 standard deviations or more, the difference is trivial, even if it is statistically significant(16). It was cognizant that, Pearson r correlation with its 95% CI was used to assess the correlation of the direct and indirect constructs of IBM. The Pearson r correlation has a value ranges from -1 to 1:+ 1(strong),+0.5(positive correlation), 0 (weak correlation) and -1(strong negative correlation), -0.5 (negative correlation)(17).