Page 78 - GHES-1-1

Global Health Econ Sustain Clinician’s attitude toward health extension program

the response rates for each item were examined using mean Cronbach’s alpha coefficient and was tested for both the
and median summaries. Two statistical models were used: subscales/factors which emerged from item reduction
factor analysis to determine latent variables and linear and for the attitude scale items as a whole.
regression analysis to determine associated factors. The (b) Outcome variables
attitude item rating scale had to be converted to run factor A total of 54-item questions were used to derive the
analysis. As a result, the attitude rating scale was recoded number of latent variables produced using EFA, which
into two scales to align with the knowledge scales. To were used to determine the outcome variables. Three
demonstrate incorrect perspectives, responses with neutral, outcome variables—the clinicians’ perceived attitude
disagree, and strongly disagree were changed to zero and toward the skill of HEWs (F1), clinicians’ knowledge
denoted as “disagree;” conversely, responses with agree towards HEP activities (F2), and clinicians’ perceived
and strongly agree were changed to one and denoted as attitude towards the impact of HEP (F3) were created
“agree.” Items were normally distributed, and due to a lack through the use of EFA. Predicted continuous values
of standards, each item question in the latent variable was were generated for each latent variable, and each factor
added up to create composite scores ranging from 0 to 54. underwent a linear regression analysis.
A higher score denotes that the clinician is knowledgeable (c) Linear regression analysis
or has a positive attitude toward HEP. The mean and Predicted values for each constructed factor with
median values of each latent variable were calculated. The enough items were generated. However, a factor
central message of the item questions served as the basis with inadequate items (≤2) was excluded from
for labeling the latent variables. Latent variables were further analysis (Yong & Pearce, 2013). An analysis
of variance (ANOVA) was used to determine the
converted into dichotomous variable types using the mean level of association between dependent variables
as a cutoff point, with values above the mean labeled as and predictors. Eleven independent variables,
“good knowledge” or “favorable attitude” and scores below of which two of them (age and experience) were
the mean labeled as “poor knowledge” or “unfavorable continuous-discrete, were included in the regression
attitude.” The factor analysis and linear regression are model. Categorical variables such as sex, marital
described as follows:
status, educational level, qualification, facility type,
(a) Factor analysis involvement in HEP outreach, involvement in HEP
The similarity between items was assessed using the review meetings, involvement in HEP home visits,
average inter-item correlation, and a diagonal correlation and willingness to work in HEP were included.
matrix was calculated to check communalities. Sampling A normality test was run before multiple linear
adequacy for both individual items and factors was regression was carried out. Multicollinearity was
measured using the Kaiser–Meyer–Olkin (KMO) checked using variance inflated factor (VIF), where
test, with values >0.6 considered adequate (Robson & VIF >10% indicates the presence of collinearity.
Haddad, 2012; VanSickle et al., 2016). Before performing An overall goodness-of-fit test was measured using
the exploratory factor analysis (EFA), three criteria were adjusted R-squared (r ). A variable with P < 0.25 in
2
checked: sample size, factorability of the correlation the one-way ANOVA test was considered statistically
matrix, and KMO measure of sampling adequacy or significant and re-entered into the multivariant linear
Bartlett’s test of sphericity. EFA was then applied, and regression model to exclude confounders. In the
the number of factors extracted from item questions multivariant linear regression model, a P < 0.05 at a
was determined using the Eigenvalues matrix, where a 95% CI was considered statistically significant.
value >1 was considered to construct factors. A scree
plot was created to determine the relative importance of 3. Results
retained factors by examining significant breaks among
dotted lines in the graph (Ul Hadia et al., 2016). The 3.1. Characteristics of respondents
total variance was used to explain factors removed due A total of 1210 clinicians were interviewed; details are
to significant breaks. Factor loading was carried out, shown in Table 1. Over half (53.4%) of the population was
and items with weak loadings (<0.4) or cross-loading in the 25- to 29-year-old age range, with a mean age of
on several factors were deleted (Winters et al., 2016). 28.3 (SD = 5.4) years. Male respondents comprised 50.8%,
Interpretation of factors was made after factor rotation and more than half (51.8%) spent the first 15 years of
to create cluster variables. The orthogonal varimax- their lives growing up in rural areas. Married respondents
type rotation method was applied to summarize the account for 621 (51.3%), with the majority (42.8%) of them
dimension of the scale (Yong & Pearce, 2013). The meeting the qualification of clinical nurses. The majority
measure of internal reliability was assessed using of clinicians (53.1%) had <5 years of experience, with an

Volume 1 Issue 1 (2023) 4 https://doi.org/10.36922/ghes.0887