A Donabedian-based instrument for measuring parental satisfaction in pediatric dental care: scale development and validation

2.1 Study design

This study was designed as a methodological study aimed at the development and validation of a parental satisfaction scale. Accordingly, established instrument development protocols were followed, including conceptualization, item generation, expert evaluation, and psychometric testing. The evaluation of the scale’s measurement properties was guided by COSMIN (COnsensus-based Standards for the selection of health Measurement INstruments) standards [20]. The workflow of the study is presented in Fig. 1.

Fig. 1.Flow diagram of the study. AVE: average variance extracted; HTMT: heterotrait-monotrait ratio.

Ethical approval was obtained from the institutional ethics committee prior to study initiation (No.: 23-5.1T/14). Written informed consent was obtained from all participating parents or legal guardians.

2.2 Phase 1: item generation

To initiate the scale development process, a comprehensive literature review was conducted to identify existing instruments assessing satisfaction in healthcare settings. A literature search was conducted in PubMed for studies published between January 2003 and January 2023 on parental satisfaction in pediatric dental services, using keywords such as “pediatric dentistry”, “parental satisfaction”, “quality of care”, “scale development”, and “psychometric validation”. The review aimed to examine conceptual frameworks, item content, and psychometric limitations of previously published scales, thereby informing the theoretical foundation of the present instrument.

Guided by the Donabedian Healthcare Quality Model, the research team generated an initial pool of 40 items. The expert panel consisted of three specialists in pediatric dentistry, one public health specialist, and one health service researcher; detailed characteristics of the experts are provided in Supplementary Table 1.

During each meeting, items were evaluated based on three criteria: (1) relevance to the construct of parental satisfaction in pediatric dental care, (2) clarity and comprehensibility for the target population, and (3) conceptual alignment with the Donabedian structure–process–outcome framework.

Item generation was conducted through two structured consensus meetings, integrating expert panel discussions and iterative brainstorming during which proposed items were systematically evaluated for relevance, clarity, and conceptual alignment with the Donabedian framework (Supplementary Table 2). The preliminary version of the scale was organized into three hypothesized domains: structure, process, and outcome.

Following expert review and content refinement, item reduction was guided by predefined conceptual and methodological criteria. Specifically, items were excluded if they (1) demonstrated semantic redundancy or conceptual overlap with other items, (2) lacked clarity or were prone to ambiguous interpretation, or (3) were considered too context-specific, thereby limiting their generalizability across different clinical settings. This process resulted in a reduced draft instrument comprising 32 items. A sociodemographic section was also incorporated to facilitate comparison between groups.

2.3 Phase 2: content validity

To establish content validity, the preliminary item pool was subjected to expert evaluation focusing on relevance, clarity, and representativeness of the construct of parental satisfaction in pediatric dental care. Content validity was ensured through a structured, expert-based consensus process; however, quantitative content validity indices were not used. Consensus was defined as agreement among at least four of the five panel members. Items that did not reach consensus were further discussed and re-evaluated in subsequent rounds.

Each item was independently assessed for its alignment with the theoretical foundations of the Donabedian Health Care Quality Model and its appropriateness for the target population. Through qualitative review and structured feedback, experts provided recommendations for revising, merging, or removing items that were redundant, ambiguous, or conceptually misaligned. Based on consensus-based qualitative judgments, the instrument was refined and finalized as a 30-item scale (Supplementary Table 2).

2.4 Phase 3: data collection and participants

The finalized version of the scale was administered between March 2023 and September 2023 at the Department of Pediatric Dentistry, Faculty of Dentistry, Ege University. Prior to the beginning of this study, a pilot test was conducted with 15 parents to assess the readability and clarity of the questionnaire items. Data collection was conducted using a cross-sectional design involving two independent samples to support the psychometric evaluation of the instrument. According to psychometric guidelines, when the relationships among items have not yet been empirically established, exploratory factor analysis (EFA) is recommended as the initial step to identify the underlying latent structure [21]. The use of confirmatory factor analysis (CFA) for exploratory purposes may be restrictive and potentially misleading [22]. Furthermore, scale development guidelines recommend the use of independent samples for EFA and CFA as a methodological standard to ensure appropriate validation of the factor structure [23, 24].

The first sample was utilized for EFA to identify the underlying factor structure. Following initial EFA, items with low loadings (<0.30) or substantial cross-loadings were removed to improve construct clarity. A second EFA was conducted to confirm the refined structure. This final structure was then tested using CFA on an independent sample. Modification indices were considered only when theoretically meaningful. Correlated errors were allowed between items measuring highly similar content, consistent with the conceptual framework. Accordingly, in the present study, two independent samples were used (n = 229 for EFA and n = 213 for CFA), adhering to this methodological standard.

Recommended sample sizes for EFA generally range from five to ten participants per item, with a minimum overall sample size of 100 being acceptable. Larger samples, around 200–300, are considered adequate or good according to established guidelines [25, 26, 27].

Although there is no universally accepted sample size for CFA, there is debate regarding using the same sample for both EFA and CFA. Methodological recommendations advise against performing CFA on the same dataset used for EFA [28]. Using the recommended participant-to-item ratios of ≥5:1 [24] and ideally 10:1 [29], the purposive CFA sample of 213 parents, corresponding to approximately 7 participants per item, was considered adequate.

Parents or legal guardians aged 18 years or older who had a child aged 0–13 years receiving dental care at the Department of Pediatric Dentistry during the study period, who were the primary decision-makers for their child’s dental care, had attended at least one dental visit prior to completing the questionnaire, were able to communicate in Turkish, and provided written informed consent were included in the study. To maintain sample independence, no participant overlapped in both the EFA and CFA datasets. The EFA and CFA samples were recruited from the same university-based pediatric dental clinic using a consecutive sampling approach during distinct time periods. The response rate was 88.1% for the EFA sample and 81.9% for the CFA sample. Participants were enrolled sequentially rather than randomly assigned.

2.5 Phase 4: assessment of reliability, validity, and scoring

Responses were recorded on a 5-point Likert scale ranging from 1 (“strongly disagree”) to 5 (“strongly agree”). The total scale score ranged from 30 to 150, with lower scores indicating lower levels of parental satisfaction.

Internal consistency reliability was assessed using Cronbach’s alpha coefficients. Construct validity was examined through exploratory and confirmatory factor analyses, and the dimensional structure was further evaluated using a second-order CFA to assess whether first-order factors loaded onto an overarching parental satisfaction construct.

2.6 Data analysis

All analyses were conducted using the R programming language and the RStudio development environment. For exploratory factor analysis, the psych [30] and GPArotation packages [31] were used. Confirmatory factor analysis and the second-order factor model were performed using the lavaan [32] and semTools [33] packages. Additionally, the semPlot [34] package was employed for model visualization.

Sampling adequacy was evaluated using the Kaiser-Meyer-Olkin (KMO) measure. The factorability of the correlation matrix was assessed using Bartlett’s Test of Sphericity. A KMO value greater than 0.60 and a statistically significant Bartlett’s test were considered indicative of adequate sampling adequacy and factorability [35, 36].

The number of factors was determined using parallel analysis, based on the comparison of observed eigenvalues with those obtained from randomly generated data. EFA was conducted using maximum likelihood estimation with oblimin rotation, given the expected correlations among factors.

CFA was performed using the Weighted Least Squares Mean and Variance adjusted estimator (WLSMV), which is appropriate for ordinal data. Model fit was evaluated using multiple indices, including the chi-square statistic (χ²), Comparative Fit Index (CFI), Tucker-Lewis Index (TLI), Root Mean Square Error of Approximation (RMSEA), and Standardized Root Mean Square Residual (SRMR). Model fit was evaluated based on commonly recommended criteria [37, 38]. While RMSEA values ≤0.06 indicate a close fit, values up to 0.08 were considered acceptable in line with previous literature.

Reliability was assessed using Cronbach’s alpha and McDonald’s omega coefficients. Convergent validity was evaluated using composite reliability (CR) and average variance extracted (AVE), while discriminant validity was assessed using the heterotrait-monotrait ratio (HTMT) and the Fornell-Larcker criterion.

3.1 Exploratory factor analysis (EFA)

The 30-item scale demonstrated excellent internal consistency (Cronbach’s alpha = 0.97), with item-total correlations ranging from 0.51 to 0.84. Removal of individual items did not improve the alpha coefficient (Table 2).

Parallel analysis was conducted to determine the number of factors. The first four observed eigenvalues (16.98, 1.58, 0.83, and 0.75) exceeded those obtained from randomly generated data (0.89, 0.66, 0.58, and 0.52). Although the fifth observed eigenvalue (0.468) slightly exceeded the mean eigenvalue obtained from simulated data (0.456), it did not surpass the more conservative 95th percentile threshold recommended for parallel analysis. These findings supported the retention of a four-factor structure. The scree plot of the parallel analysis is presented in Fig. 2.

Fig. 2.Parallel analysis scree plot.

A four-factor solution was specified and estimated using maximum likelihood extraction with oblimin rotation, given the expected correlations among factors. Factor loadings were examined to evaluate the structure and the distribution of items across the factors (Table 3).

In this study, loadings of 0.30 and above were regarded as meaningful. Higher factor loadings indicate that items are more strongly associated with their respective factors and contribute to well-defined latent constructs. Initial EFA results revealed that some items exhibited cross-loadings. These findings suggested that these items did not clearly represent a single factor; therefore, item reduction was applied.

Item reduction was guided by both statistical criteria and conceptual evaluation. Items I3 and I4, related to the adequacy of dentists and support staff, were excluded due to cross-loadings. Although intended to reflect structural aspects of care, these items appeared to overlap with the “Accessibility” domain, suggesting that parents may associate the number of dentists/support staff with ease of access to care. Item I15 was removed due to an unstable loading. As this item referred to a hypothetical situation, participants who had not encountered any problems may have responded based on general expectations rather than actual experiences, potentially introducing ambiguity in interpretation. Items I21 and I22, both addressing appointment scheduling, showed high loadings but reflected a narrow and specific aspect of service delivery. To enhance the parsimony and conceptual clarity of the scale, the more general item (I22) was retained, and the more specific item (I21) was removed. Finally, item I26 was excluded due to cross-loading, as it represented a global evaluation of care and overlapped with multiple domains.

Based on the criteria of cross-loading, low factor loading, and item redundancy, items I3, I4, I15, I26, I20, and I21 were excluded from the scale. Following item removal (Table 4), the factor structure became clearer and more parsimonious, with substantially reduced cross-loadings and items primarily loading on their intended factors. This pattern is further illustrated in the heatmap (Fig. 3). Factor loadings showed distinct clustering, indicating well-defined and internally consistent dimensions, although minor secondary loadings suggest the presence of related components. While the first factor explained the largest share of variance, the remaining factors contributed meaningfully, supporting a multidimensional structure suitable for subsequent confirmatory analysis.

Fig. 3.Heatmap of exploratory factor analysis loadings after Item reduction. ML: maximum likelihood.

Examination of cross-loadings showed that some items had minor loadings on additional factors; however, as these values were below 0.30 and each item demonstrated a dominant loading on a single factor, they were not considered to represent meaningful cross-loading.

The proportion of variance explained by each factor was calculated. The four-factor structure accounted for 71% of the total variance, with the first factor explaining the largest share (28%), followed by the second (16%), third (14%), and fourth (13%) factors (Supplementary Table 3). These findings indicate that all factors contributed meaningfully to the overall variance and support the adequacy of the multidimensional structure of the scale.

3.2 Confirmatory factor analysis (CFA)

Confirmatory factor analysis (CFA) was conducted to evaluate the fit of the four-factor structure identified in the EFA. The analysis was performed on the second independent sample (N = 213), yielding an approximate participant-to-item ratio of 7:1, consistent with recommended guidelines for structural equation modeling [38, 41].

Parameter estimation was carried out using the WLSMV estimator in R, which is recommended for ordinal data [21, 42]. This estimator accounts for non-normality and provides robust standard errors and adjusted chi-square statistics [43]. The four-factor model identified in the exploratory phase was specified a priori. The latent variables and their respective observed indicators were as follows:

● Professional Care & Communication (Factor 1): Items 1, 2, 5–13.

● Service Delivery (Factor 2): Items 14, 16–19.

● Outcome (Factor 3): Items 27–30.

● Accessibility (Factor 4): Items 22–25.

The model fit indices indicated that χ²(df = 246) = 695.07 (p < 0.001), CFI = 0.974, TLI = 0.971, RMSEA = 0.093 (90% Confidence Interval (CI): 0.085, 0.101), and SRMR = 0.046. Although the initial CFA demonstrated good fit according to CFI and TLI, the RMSEA value exceeded the recommended threshold [37], which may be attributed to shared variance among conceptually similar items not fully captured by the latent structure.

Based on modification indices and guided by theoretical considerations, a number of cross-loadings were introduced to improve model performance (Table 5).

Modification indices were examined to identify parameters with the potential to improve model fit. The inclusion of theoretically justified error covariances can be interpreted as accounting for local dependence arising from conceptually similar items. Such modifications are common in scale development when items share similar wording or capture closely related aspects of a construct. Only theoretically supported modifications were retained, while cross-loading or conceptually unjustified changes were excluded to minimize the risk of overfitting.

The highest modification indices were observed between items I5–I6, I7–I8, and I29–I30. The inclusion of correlated error terms was theoretically justified based on the conceptual similarity of specific item pairs. The covariance between items I5 and I6 reflects a shared underlying theme of trust and perceived safety, as I5 assesses trust in the dentist’s professional competence, while I6 captures the patient’s sense of safety during treatment. Similarly, items I7 and I8 both relate to the dentist’s interpersonal behavior, with I7 focusing on child and I8 on the parent, indicating provider–patient interaction. Furthermore, the covariance between items I29 and I30 is supported by their shared focus on behavioral intentions as outcomes of satisfaction. These items represent closely related outcome-oriented evaluations of care. Taken together, these modifications were considered theoretically meaningful and consistent with the underlying construct, while also contributing to improved model fit.

Following these modifications, a revised CFA model was estimated, and the corresponding fit indices were obtained. The revised model demonstrated improved fit compared to the initial model, as reflected by increased CFI (0.982) and TLI (0.980) values and decreased RMSEA (0.078) and SRMR (0.042). The CFI and TLI values exceeded the 0.95 threshold, indicating a good model fit, while the RMSEA value suggested an acceptable level of fit. The SRMR value further supported good model fit by indicating a low residual error. Overall, these results indicate that the applied modifications led to a meaningful improvement in model fit.

The standardized factor loadings of the revised model are presented in Table 6 and visually summarized in the heatmap (Fig. 4).

Fig. 4.Heatmap of standardized factor loadings (Revised CFA Model).

After these modifications, every item loaded significantly onto its designated factor. Given the well-documented sensitivity of the chi-square statistic to sample size [38, 41], model evaluation focused primarily on relative and incremental fit indices.

Examination of the standardized factor loadings indicated that all items loaded strongly on their respective factors, with values ranging from 0.65 to 0.95. All loadings exceeded the commonly accepted threshold of 0.50, supporting good construct validity. Particularly high loadings were observed for items I25, I28, and I29, indicating that these items strongly represent their underlying latent constructs. In particular, positive perceptions of service accessibility (I25) may contribute overall satisfaction (I28), which in turn is strongly linked to future behavioral intentions such as willingness to use the services again (I29). Therefore, the observed overlap can be interpreted as a theoretically coherent pattern consistent with the multidimensional yet interrelated nature of healthcare service evaluation, rather than mere item redundancy.

3.3 Internal consistency and reliability

Reliability analyses based on Cronbach’s alpha and McDonald’s omega coefficients were conducted to assess the internal consistency of the final scale. The results are presented in Table 7.

Cronbach’s alpha and McDonald’s omega indicated good to excellent internal consistency across subscales (α = 0.81–0.94; ω = 0.88–0.96), with high overall reliability (α = 0.95; ω = 0.97). Although such high values may suggest item redundancy, the multidimensional structure supports a coherent rather than overlapping construct. Convergent validity was supported by acceptable CR and AVE values (Supplementary Table 4). Discriminant validity was also confirmed, as all HTMT values were below the recommended threshold of 0.85 (Supplementary Table 5) [35, 44, 45]. Following the evaluation of reliability and validity, inter-factor correlations of the final model were examined to further explore the relationships among the constructs (Table 8).

Examination of inter-factor correlations revealed moderate to high relationships among the factors. Some correlations exceeded 0.70, indicating that the factors are related, while all values remained below 0.90, suggesting that the constructs are distinguishable rather than redundant.

Discriminant validity was further evaluated using the Fornell-Larcker criterion, which requires that the square root of AVE for each construct exceeds its correlations with other constructs [44]. This condition was met for most factors, with the square root of AVE values (0.86 for F1, 0.87 for F2, and 0.93 for F3) exceeding the corresponding inter-factor correlations (0.66–0.81). However, for F4, the square root of AVE (0.81) was very close to its correlation with F3 (0.81), indicating a high degree of overlap between these two constructs.

Overall, these findings support acceptable discriminant validity, while also suggesting that the relatively high inter-factor correlations may reflect the presence of a broader underlying construct. Accordingly, a second-order confirmatory factor analysis was conducted to examine the hierarchical structure of the scale. The results are presented in Table 9, and model fit indices were reported based on scaled estimates obtained using the Diagonally Weighted Least Squares (DWLS) estimator in R.

Both models demonstrated good fit to the data. Examination of the goodness-of-fit indices for the second-order model showed that the CFI (0.977) and TLI (0.974) values exceeded the critical threshold of 0.95, indicating excellent model fit.

The RMSEA value was 0.088 (90% CI [0.079–0.096]). Although values ≤0.06 are often considered indicative of close fit, values up to 0.08–0.10 may be acceptable depending on model complexity and should be interpreted alongside other fit indices. In the present study, the high CFI and TLI values support the overall adequacy of model fit despite the slightly elevated RMSEA [37, 38, 46, 47, 48, 49].

Second-order CFA showed that the general satisfaction factor loaded strongly on all subdimensions (0.81–0.92; p < 0.001), indicating that the higher-order construct is well represented by the underlying factors (Supplementary Table 6). These findings support a multidimensional yet integrated structure and the appropriateness of a hierarchical model (Fig. 5).

Fig. 5.Path diagram of the second-order confirmatory factor analysis.

To assess the extent to which the second-order structure explains relationships among first-order dimensions, the target coefficient (T) was calculated using scaled chi-square values [49]. The resulting value (T = 0.87) indicates that the higher-order factor accounts for a substantial proportion of covariance among the first-order factors. Consistently, the higher-order factor explained a large proportion of variance in each dimension (66–85%).

Together with high inter-factor correlations and strong factor loadings, these findings support the adequacy and parsimony of a hierarchical model over a bifactor alternative.

3.4 Scoring guide

The scale used in this study was rated on a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). Higher scores indicate higher levels of satisfaction. Subscale scores were calculated using a unit-weighting approach, whereby the mean of the items corresponding to each factor was computed. Similarly, the total score was obtained by averaging all items.

The use of equal weighting (i.e., summing or averaging items) has been recommended as a reliable and robust approach in the literature, as it reduces the influence of sample-specific factor loadings and enhances the generalizability of the scale across different samples [50]. Given the high and relatively homogeneous factor loadings observed in this study, the unit-weighting method was deemed appropriate. Since all items were scored in the same direction, no reverse coding was required.