Contingency theory holds that no single costing system design suits all organisations; rather, the appropriate level of CSS depends on the surrounding context ([3]; [10]; [33]). [22] distinguish a selection form of fit, which tests only whether contextual factors are associated with system design, from a mediation form, which additionally tests whether system design in turn affects performance. [20] argued that most contingency based management accounting research has relied on the simpler selection form, leaving the field able to describe whether a costing system matches its context without saying whether that match delivers any benefit. The present study follows the mediation approach, on the reasoning that a contextual factor which predicts CSS is of limited practical interest unless CSS itself is subsequently shown to affect outcomes managers care about. Contextual factors are therefore specified as antecedents of CSS (Hypotheses 1 - 6), and CSS as an antecedent of non financial performance, hypothesised to mediate its effect on financial performance (Hypotheses 7 - 9).

Competition is expected to increase the value of accurate cost information for pricing and process decisions, since competitive markets punish pricing errors more severely ([12]; [26]). Evidence is mixed: [38] found a significant effect while [17], [27], and [14] did not; [26] critique of unidimensional competition measures offers one candidate explanation, addressed here through a multi item scale, though this cannot be confirmed without direct replication.

H1:Competition is positively related to CSS.

Service diversity and differentiation strategy are expected to raise the complexity of resource consumption across services; differentiation strategy is additionally expected to raise service diversity and indirect cost, since differentiation oriented firms typically invest more in service variety and innovation ([39]), so its effect on CSS may operate partly indirectly.

H2a-c:Service diversity is positively related to CSS (a) and to cost structure (b); cost structure positively mediates the diversity-CSS relationship (c).

H3a-e: Differentiation strategy is positively related to CSS (a) and to service diversity (b); service diversity positively mediates the differentiation-CSS relationship (c); differentiation strategy is positively related to cost structure (d); and cost structure positively mediates the differentiation-CSS relationship (e).

Cost leadership strategy is hypothesised to relate negatively to CSS, on the reasoning that cost leadership firms already understand their cost base well, reducing the incremental value of further sophistication ([17]). This reasoning is contested rather than settled: a firm competing on cost might equally invest more, not less, in the cost visibility CSS provides, since small errors are more consequential to a lowest-cost competitor. The hypothesis below follows the majority position in the literature while acknowledging this competing logic.

H4:Cost leadership strategy is negatively related to CSS.

Cost structure is expected to increase CSS directly, since higher indirect cost intensity raises the cost of cross-subsidisation errors under a simpler system ([11]; [13]). Organisational size is expected to increase CSS via greater administrative and financial resources ([4]; [38]), although [23] recent evidence that size’s effect is conditional on firm age and mediated by product diversity suggests the direct-effect hypothesis tested here is a simplification.

H5:Cost structure is positively related to CSS.

H6:Organisational size is positively related to CSS.

CSS is expected to improve service quality and reduce service cycle time by giving managers more accurate, activity level information ([27]; [28]), supporting cost reduction and financial performance ([11]; [30]; [31]). Because prior ABC adoption research has largely failed to find a direct financial performance effect ([11]; [28]; [31]), the CSS-financial performance relationship is treated as potentially both direct and mediated, rather than assumed to be purely one or the other.

H7a, b, c:CSS is positively related to service quality (a); service quality positively mediates CSS’s effect on financial performance (b); service quality is positively related to service cycle time reduction (c).

Hypothesis 7d, that service quality is negatively related to cost reduction, requires separate justification because the direction runs counter to the generally positive tenor of the other hypotheses; this direction was specified a priori, before estimation, not adjusted afterwards to fit a result. The logic is a quality-cost trade-off specific to labour intensive services: sustaining higher perceived service quality typically requires additional staff input or lower staff-to-client ratios, which raises rather than reduces cost ([36])—unlike manufacturing, where quality improvements (fewer defects) often cut cost by reducing rework, the setting in which most positive quality-cost findings were established. Because cost reduction is coded so higher scores indicate greater perceived cost reduction, a negative coefficient is consistent with, not contrary to, this trade-off. The interpretation is theoretically defensible given the service context but remains an ex ante hypothesis under test, revisited critically in the discussion.

H7d:Service quality is negatively related to cost reduction.

H8a, b, c:CSS is positively related to service cycle time reduction (a); service cycle time reduction positively mediates CSS’s effect on financial performance (b) and is positively related to cost reduction (c).

H9a, b:CSS is positively related to cost reduction (a); cost reduction positively mediates CSS’s effect on financial performance (b).

Figure 1 summarises the hypothesised model. Solid lines indicate a hypothesised direct relationship; dotted lines indicate a hypothesised mediation relationship. This is the a priori structure; Section 5.2 reports whether the estimated model required any modification.

Figure 1.The hypothesised CSS research model.

A cross-sectional survey questionnaire was used. The population comprised medium and large UK non-manufacturing firms (≥50 employees, ≥£25 million annual sales), sampled from the FAME database across thirteen UK Standard Industrial Classification service sections. From a sampling frame of 2969 eligible firms, 2000 firms were randomly selected and one questionnaire was sent to each selected firm. Management accountants or financial directors were targeted because they were expected to have the most relevant knowledge of costing practices. Where a large firm contained several divisions, respondents were instructed to answer from the perspective of the business unit that most clearly defined where they worked. Thus, the unit of analysis is the business unit represented by each returned questionnaire, with one retained questionnaire constituting one independent observational unit. Of 263 returned questionnaires (13.15%), 219 were usable (10.95% effective response rate). Of these, 204 business units reported an operating costing system and were retained for the CSS model; the remaining 15 units reported no costing system and were excluded because CSS, defined by costing-system design characteristics, was not meaningful for those units.

CSS was operationalised using three observed design characteristics: the number of cost pools, the number of volume cost drivers, and the number of non-volume cost drivers. This specification follows the study’s original SEM model and treats these indicators reflectively, on the premise that a common underlying level of costing-system sophistication is manifested in greater differentiation of cost pools and greater use of both volume and non-volume drivers. We recognise that the indicators are distinct design features and could alternatively be conceptualised as formative components. Accordingly, the reflective specification is retained to preserve consistency with the prespecified model and original analysis, but this alternative formative interpretation is acknowledged as a measurement limitation. Business units using variable costing systems were not excluded; for these units the number of cost pools used for product/service costing was recoded to zero. Because the raw count indicators were highly skewed and contained many zeros, the original analysis used log-transformed values for the SEM/CFA; raw values are reported descriptively in Section 5.1. The final standardised CFA loadings for CSS were 0.64 (cost pools), 0.91 (volume cost drivers), and 0.84 (non-volume cost drivers).

Competition, service diversity, differentiation strategy, and cost leadership strategy were each measured with multi item five point Likert scales adapted from prior contingency based costing research and validated by CFA; competition used a multi item scale reflecting a broad conceptualisation of competitive forces ([35]) rather than a single price-competition item, following [26] critique of unidimensional competition measures. Differentiation strategy retained only two items after one was removed for discriminant validity reasons, a thinner basis than the other scales and a limitation acknowledged in Section 7.1. Cost structure was measured as a single indicator, the percentage of indirect to total service cost. Organisational size was measured using two indicators, log transformed employee count and sales revenue; a third indicator, capital employed, was dropped after an unacceptable factor loading of 0.21. Service quality, service cycle time reduction, cost reduction, and financial performance were each measured with multi item five point Likert scales capturing perceived improvement over the preceding period, with cost reduction coded such that higher scores indicate greater perceived cost reduction. Item counts, Cronbach’s alpha, and composite reliability for every construct are reported in full in Table 1, established once for the full set of constructs shared across this study’s costing system models using the sample of 204 business units.

Using the standard CFA reliability formula CR = (Σλ)²/[(Σλ)² + Σ(1 − λ²)], the three CSS loadings (0.64, 0.91, 0.84) yield CR = 0.84 and AVE = 0.65; Cronbach’s α was 0.82. For Size, the two retained standardised loadings (0.69 and 0.68) yield CR = 0.64 and AVE = 0.47; Cronbach’s α was 0.64. The previously reported CR values of 0.95 for CSS and 0.85 for Size were transcription/calculation errors and have been corrected. CSS also had MSV = 0.58 and ASV = 0.39.

Table 1.Construct reliability and validity for the shared measurement model (n = 204).

ᵃRetained two items after one was removed for discriminant-validity reasons; acknowledged as a measurement limitation in Section 7.1. Size’s AVE (0.47) is marginally below the conventional 0.50 threshold and its corrected CR is 0.64; these values are reported transparently and interpreted cautiously given the two-indicator specification. CR was recalculated from the final standardised CFA loadings using CR = (Σλ)²/[(Σλ)² + Σ(1 − λ²)]. All other values reflect the final measurement model after the reported CFA diagnostics.

Discriminant validity was assessed using the [21]) AVE based criterion, satisfied for all constructs after the diagnostic item removals described above, and cross-checked using the Heterotrait-Monotrait ratio (HTMT), an increasingly recommended, more sensitive alternative ([25]). As a supplementary diagnostic, the full HTMT matrix was computed for all eight reflective constructs: the highest observed ratio was 0.821 (between service diversity and service cycle time reduction), and all pairwise ratios fell below the conservative 0.85 threshold, supporting discriminant validity and reinforcing the Fornell-Larcker conclusion above. Multicollinearity among the exogenous predictors of CSS was assessed via the bivariate correlation matrix (highest correlation 0.79) and, as a further supplementary diagnostic, via Variance Inflation Factor (VIF): the maximum VIF among the six predictors of CSS was 2.66 (service diversity), well below the conventional threshold of 5, indicating no problematic multicollinearity.

Because the perceptual constructs were collected in a single questionnaire from one respondent per business unit, common method bias is a potential concern. Procedural safeguards included varying question format and item order and assuring respondents of anonymity and confidentiality ([34]). Harman’s single-factor test was recalculated using the 27 retained perceptual Likert indicators in the final model (3 competition, 3 service diversity, 2 differentiation strategy, 3 cost leadership strategy, 8 service quality, 3 service cycle-time reduction, 2 cost reduction, and 3 financial-performance items). The objective count/size indicators used for CSS and Size were not included in this common-method diagnostic. Unrotated principal-component analysis produced three components with eigenvalues greater than 1, and the first component accounted for 51.4% of total variance. Thus, no single-factor solution emerged, but the first-component share slightly exceeded the commonly used 50% caution threshold; common method bias therefore cannot be ruled out and is treated as a limitation.

Data were analysed in AMOS using a two-stage SEM procedure comprising a CFA measurement model followed by structural-model estimation. Maximum likelihood estimation (MLE) was used for the measurement parameters and latent constructs. Prior to SEM, outliers and distributional shape were examined. The raw cost-pool, cost-driver, volume-driver, non-volume-driver, employee, sales-revenue, and capital-employed variables showed substantial outliers/skewness; the original analysis therefore applied Log N transformations to these seven variables. One extreme cost-structure observation was winsorised. The transformed CSS indicators, rather than the raw zero-heavy counts, were used in CFA and SEM. Indirect effects were assessed in dedicated mediation models using 2000 bootstrap resamples and 95% confidence intervals. The structural-model and mediation results are reported separately in Section 5. Model fit was assessed using χ²/df, RMSEA and SRMR, CFI, IFI, TLI/NNFI, and PNFI.

Participation was voluntary and anonymous. A covering letter accompanying the questionnaire explained the study’s purpose, assured respondents of confidentiality, and stated that individual business units and respondents would not be identifiable in any resulting output. Ethical approval for the wider research programme of which this study forms part was obtained from the author’s institutional research ethics committee prior to data collection.

Using the raw descriptive data, the mean number of cost pools used for product/service costing was 11.37 (median 1.00) after recoding cost pools to zero for variable-costing units, and the mean number of cost drivers was 3.06. Among the 104 business units with at least one reported cost driver, the corresponding conditional mean was 6.01 drivers. Within this same 104-unit subset, the mean numbers of volume and non-volume cost drivers were 3.36 and 2.65, respectively. Across all 204 units, the corresponding unconditional means were 1.71 and 1.35. The earlier conditional means (4.76, 2.91, and 1.84) were inconsistent with the underlying dataset and have been replaced by values recalculated directly from the raw observations. These descriptive statistics are reported in Table 2.

Table 2.Descriptive statistics for the CSS indicators.

ᵃBased on all respondents (n = 204). ᵇBased only on respondents reporting cost drivers (n = 104).

CFA supported the measurement model after the standard diagnostic sequence described in Section 4.2 (factor loading review, model fit, modification indices, reliability, and discriminant validity). The structural model reported here is the a priori hypothesised model shown in Figure 1; no path was added to or removed from this model after estimation. Table 3 reports the full set of fit indices for the estimated structural model.

Table 3. Fit indices for the CSS structural model.

n = 204. Significant χ² statistics are common in models of this size and are not, on their own, treated as evidence of poor fit ([24]).

The structural model showed a mixed pattern of fit. The χ²/df, RMSEA, CFI, IFI, and PNFI values met conventional criteria, whereas SRMR (0.11) exceeded the commonly recommended 0.08 threshold and TLI/NNFI was at the boundary value of 0.90. Accordingly, the model is not characterised as demonstrating unequivocally good fit. Rather, the overall pattern suggests broadly adequate fit with evidence of residual model misspecification. No post-hoc structural paths were introduced solely to improve fit, and the findings should therefore be interpreted with appropriate caution.

This model fit assessment is based exclusively on the original structural equation modelling procedure. Hypothesis decisions throughout this paper are based exclusively on the prespecified structural model estimated using this original SEM procedure; no structural paths were added or removed after examination of the results.

Table 4 reports the standardised structural path coefficients for all hypothesised relationships.

Table 4. Structural model results: standardised path coefficients.

*p ≤ 0.05 (two-tailed); ns = not significant. n = 204. This is the a priori hypothesised model reported in full; no paths were added or removed after estimation.

Figure 2 presents the estimated model using the same layout as Figure 1: solid lines indicate paths for which a statistically significant relationship was found, and dotted lines indicate paths for which no significant relationship was found.

Figure 2.The hypotheses results for the CSS model.

Table 5 reports the mediation analyses based on 2000 bootstrap resamples and 95% confidence intervals. To avoid conflating estimates from different model specifications, Table 4 is explicitly identified as the standardised coefficients from the full structural model, while Table 5 contains standardised direct, indirect, and total effects estimated in the dedicated mediation analyses used in the original study. These mediation estimates therefore should not be reconstructed by multiplying the rounded coefficients in Table 4. Cost structure mediated the differentiation strategy-CSS relationship (indirect β = 0.07, 95% CI [0.02, 0.18]), whereas the service-diversity mediation paths were not significant. On the performance side, service quality (indirect β = 0.21, 95% CI [0.06, 0.43]) and cost reduction (indirect β = 0.30, 95% CI [0.08, 0.87]) showed significant indirect effects, while service cycle-time reduction did not. All β values in Table 4 and Table 5 are standardised; the confidence intervals in Table 5 are bootstrap intervals from the mediation analyses.

Table 5.Mediation (indirect-effect) results for key paths, CSS model.

ᵃStandardised indirect effect from the dedicated mediation analysis (not the product of rounded Table 4 coefficients). *Significant at p ≤ 0.05 (CI excludes zero). LLCI/ULCI = 95% bootstrapped confidence interval bounds (2000 resamples).

Competition was positively associated with CSS (β = 0.28, p ≤ 0.05), supporting H1. This finding is consistent with the theoretical argument that competitive pressure increases the value of more accurate and detailed cost information ([12]; [26]), and with [38]. It contrasts with [17], [27], and [14] recent Greek hotel evidence, where competition did not predict CSS; this study’s multidimensional competition measure, following [26] recommendation, or genuine sectoral variation, are both plausible, undemonstrated explanations for this divergence. However, given the cross-sectional design and the less-than-ideal overall model fit reported in Section 5.2, this finding should be interpreted as evidence of association rather than as a definitive causal relationship.

Cost structure also significantly predicted CSS (β = 0.20), consistent with the expectation that higher indirect cost intensity raises the cost of cross-subsidisation errors ([11]; [13]), and with [14]. It disagrees with several manufacturing-based studies ([4]; [17]; [27]). The most parsimonious, though unproven, explanation is sectoral: non manufacturing firms carry structurally higher indirect cost than manufacturing firms, so a relationship weak in manufacturing samples may become detectable here—an interpretation strengthened, not proven, by its convergence with [14] independent service-sector finding, and offered as one specific, falsifiable account rather than a general claim that “context matters”.

Service diversity, differentiation strategy, cost leadership strategy, and size were all unrelated to CSS directly. The differentiation strategy null result contradicts [38], and a specific measurement reason is available: only two items were retained here against six in [38] instrument, a thin basis for so multifaceted a construct; this is consistent with [18] similar null finding and [14] null business strategy finding, suggesting the difficulty may lie partly in how business strategy is operationalised across this literature generally, not only in this study’s instrument.

The size result is more interesting because it survives this measurement critique: size was measured with two well established indicators and still came back null, consistent with [27] but contradicting [38] and [4]. [23] recent evidence offers a specific, testable explanation: size’s relationship with CSS in UK manufacturing firms is conditional on firm age and operates indirectly through product diversity, not as a simple direct effect. The present study tested only a direct path, and it is plausible its null result reflects this under-specification rather than the genuine absence of any size-CSS relationship—a testable claim for future research, not a conclusion this study’s data can adjudicate. [14] contrasting significant size finding, using a similarly simple specification, further suggests that whether a direct-effect model detects a size effect may depend on unmeasured moderators that vary by setting.

CSS was positively related to service quality, service cycle time reduction, and cost reduction, and was associated with financial performance both directly and indirectly through service quality and cost reduction. Most prior ABC adoption research has failed to find a direct financial performance link ([11]; [28]; [31]), so this result deserves caution rather than being reported as a straightforward improvement; [41] similarly found that the extent of ABC use, rather than adoption per se, better predicted performance among Thai firms, lending some support to the idea that intensity measures capture variation binary adoption measures miss. The likely explanation is that CSS captures a broader range of costing practice than a binary ABC dummy, including firms with considerable sophistication never formally labelled “ABC”—a genuine advantage, but one that also means this result is not strictly comparable to null ABC findings elsewhere and should not be read as evidence CSS “works” where ABC adoption “does not”. The positive service cycle time result echoes [19] Chinese manufacturing evidence, suggesting this operational benefit may generalise across sectors, though further replication would strengthen the claim.

H7d was supported in the original structural model (β = −0.20, p ≤ 0.05), consistent with a quality-cost trade-off specific to labour intensive services (Section 3.3). The coding of the cost reduction construct was verified directly against the underlying item wording: both items ask respondents to rate the extent of cost reduction achieved, with higher scores representing greater perceived cost reduction. Because higher values of the cost-reduction construct represent greater perceived cost reduction, the negative coefficient indicates that higher service quality was associated with lower perceived cost reduction. This finding is consistent with the proposed quality-cost trade-off in labour-intensive service settings, although the cross-sectional design does not permit causal interpretation, and the trade-off itself should be understood as a plausible explanation for the observed association rather than a causally demonstrated mechanism.

This study, like most of the survey literature it draws on, uses cross-sectional, single-respondent, self-reported data. The mediation results in Table 5 are evidence of statistical association consistent with the hypothesised model, not evidence that CSS causes improved financial performance; the data cannot rule out a third factor, such as access to capital or management quality, driving both CSS and performance simultaneously. The consistency of this pattern, operational mediators mattering more than a direct link, across this study and companion analyses of ABC adoption and activity management usage in the same sample is suggestive of a plausible mechanism, but a cross-sectional design cannot demonstrate it, and no such claim is made here.

This study has several limitations. The cross-sectional design precludes causal inference; the Table 5 mediation results describe statistical patterns consistent with the hypothesised model, not demonstrated mechanisms. All study constructs were obtained from a single respondent in each business unit using a self-reported questionnaire, creating the possibility of common method bias. Harman’s single-factor analysis did not identify a single-factor structure; however, the first factor accounted for 51.6% of the total variance. Because this slightly exceeds the commonly used 50% cautionary benchmark and Harman’s test is itself a limited diagnostic, common method bias cannot be ruled out, and the results should be interpreted with this limitation in mind. Performance was measured perceptually rather than with objective accounting data. The differentiation strategy measure retained only two items, a plausible explanation for its null result. CSS was estimated reflectively to remain consistent with the prespecified original SEM, although its cost-pool and driver indicators can also be viewed as formative design components; future research should compare reflective and formative/composite specifications directly. The study is confined to the UK non manufacturing sector, limiting generalisability to other countries and to manufacturing; direct manufacturing-services comparison within one design would test the sectoral explanations in Section 6 more directly than between-study comparison can. Finally, the model omits variables recent evidence suggests may matter, notably firm age as a boundary condition on size ([23]) and service complexity as an alternative to size, so the size null result should be read as evidence against this specific model rather than a settled conclusion.

Future research should prioritise longitudinal or matched-pair, multiple-respondent designs; objective financial performance measures; cross-country replication and direct manufacturing-services comparison within one design; a refined differentiation strategy scale; explicit tests of firm age and service complexity following [23]; and attention to how increasingly digitalised costing-system characteristics, not captured by this study’s cost-pool/driver measure, might extend the constructs used throughout this literature.