nomo_reliability() estimates score reliability from a fitted first-order
CFA measurement model. The primary estimate is a model-based omega-type
composite reliability from semTools::compRelSEM(). Coefficient alpha is
available as a familiar secondary statistic and is explicitly qualified by
its stronger assumptions.
Usage
nomo_reliability(
fit,
obs.var = TRUE,
ordinal_scale = TRUE,
include_alpha = TRUE,
ci = c("none", "bootstrap"),
ci_level = 0.95,
ci_boot = 1000L,
ci_seed = NULL,
ci_ncpus = 1L,
guidance = nomo_defaults()
)Arguments
- fit
A fitted
nomo_cfa()object or a fittedlavaanCFA model.- obs.var
Logical passed to
semTools::compRelSEM().TRUE(default) uses observed covariances in the denominator;FALSEuses model-implied covariances.- ordinal_scale
Logical passed as
ord.scaletosemTools::compRelSEM(). For an all-ordered composite,TRUE(default) estimates reliability on the actual ordinal-score scale using the Green-Yang correction implemented bysemTools;FALSEestimates the latent-response-scale coefficient.- include_alpha
Logical. If
TRUE(default), also return coefficient alpha as a secondary statistic where the requested estimand is defined. For ordered indicators withordinal_scale = TRUE, observed-score alpha is deliberately reported as unavailable rather than silently switching to latent-response ("ordinal alpha") or numeric-score alpha. Alpha is not treated as the preferred reliability estimate for a general congeneric CFA.- ci
Character.
"none"(default) returns point estimates only."bootstrap"adds nonparametric percentile confidence intervals by repeatedly refitting the same CFA withlavaan::bootstrapLavaan().- ci_level
Confidence level for bootstrap intervals. Default is
0.95.- ci_boot
Number of ordinary bootstrap resamples when
ci = "bootstrap". Default is1000.- ci_seed
Optional integer seed for reproducible bootstrap intervals.
- ci_ncpus
Number of worker processes for the bootstrap. The default,
1, runs serially. Values above one use lavaan'ssnowbackend, which works on Windows, macOS, and Linux. The draws depend on the worker count as well as the seed, so a result is reproducible for a given seed and worker count, and both are recorded. More workers are not always faster: each worker must start and load its packages, and in the evaluation recorded in the package's issue tracker, eight workers were slower than four. Workers must be able to load nomologR, lavaan, and semTools from the library.- guidance
Guidance settings from
nomo_defaults(). The configured reliability reference is a review prompt, not a pass/fail criterion.
Value
A nomo_reliability object. The fields to read are:
omega: model-based omega per construct, with bootstrap intervals whenci = "bootstrap".alpha: coefficient alpha, reported as secondary.evidence: both coefficients with their references and interpretations.alpha_statusandci_status: what was computed, and why anything was not.model_strainandimproper_solution: whether the measurement model showed strain that qualifies the coefficients.referencesanddecision_log.
attention in evidence and severity in decision_log are "info",
"review", or "concern", as is signal in the "coefficients" table
of nomo_table(). See Conventions in returned tables in ?nomologR.
Other fields record the call, the settings used, and intermediate engine
results. They may change between releases and are not part of the stable
interface (see ?nomologR).
print() shows the range of omega, the flags it raised against the review
reference, and how uncertainty was handled; summary() adds each
construct's omega and alpha with their intervals and score scales, the
reason any alpha was not computed, and every flag in the decision log
with its reason.
Details
The function is deliberately measurement-first: it refuses nonconverged models, structural SEMs, higher-order models, and cross-loaded first-order indicators. When global CFA strain or an improper solution is present, reliability is still inspectable but the decision log warns that model-based reliability can be distorted by misspecification.
For ordered indicators, the observed-score omega (ordinal_scale = TRUE)
is the same whether lavaan fitted the model with its delta or theta
parameterization. The latent-response coefficients (ordinal_scale = FALSE) are computed in the model's unstandardized metric and differ between
the two; with a theta-parameterized fit the decision log says so.
Bootstrap intervals refit the model to resampled cases, so they need the
raw data: a model fitted from a covariance matrix has none, and its
ci_status says so.
Average variance extracted (AVE) is not a reliability coefficient and is
intentionally handled by nomo_validity() instead.
References
Historical context:
Cronbach, L. J. (1951). Coefficient alpha and the internal structure of tests. Psychometrika, 16(3), 297-334. doi:10.1007/BF02310555
Contemporary model-based reliability:
Bell, S. M., Chalmers, R. P., & Flora, D. B. (2024). The impact of measurement model misspecification on coefficient omega estimates of composite reliability. Educational and Psychological Measurement, 84(1), 5-39. doi:10.1177/00131644231155804
Dunn, T. J., Baguley, T., & Brunsden, V. (2014). From alpha to omega: A practical solution to the pervasive problem of internal consistency estimation. British Journal of Psychology, 105(3), 399-412. doi:10.1111/bjop.12046
Flora, D. B. (2020). Your coefficient alpha is probably wrong, but which coefficient omega is right? A tutorial on using R to obtain better reliability estimates. Advances in Methods and Practices in Psychological Science, 3(4), 484-501. doi:10.1177/2515245920951747
Green, S. B., & Yang, Y. (2009). Reliability of summed item scores using structural equation modeling: An alternative to coefficient alpha. Psychometrika, 74(1), 155-167. doi:10.1007/s11336-008-9099-3
Kelley, K., & Pornprasertmanit, S. (2016). Confidence intervals for population reliability coefficients: Evaluation of methods, recommendations, and software for composite measures. Psychological Methods, 21(1), 69-92. doi:10.1037/a0040086
McNeish, D. (2018). Thanks coefficient alpha, we'll take it from here. Psychological Methods, 23(3), 412-433. doi:10.1037/met0000144
Sijtsma, K. (2009). On the use, the misuse, and the very limited usefulness of Cronbach's alpha. Psychometrika, 74(1), 107-120. doi:10.1007/s11336-008-9101-0
Examples
model <- '
visual =~ x1 + x2 + x3
textual =~ x4 + x5 + x6
speed =~ x7 + x8 + x9
'
cfa <- nomo_cfa(model, data = lavaan::HolzingerSwineford1939)
rel <- nomo_reliability(cfa)
summary(rel)
#> <nomo_reliability summary> Reliability
#> Constructs: 3 | Review reference: .70
#>
#> Coefficients
#> Construct Indicators Omega Alpha Flag
#> visual continuous .61 .63 Review
#> textual continuous .89 .88
#> speed continuous .69 .69 Review
#>
#> Flagged
#> - visual (Review): Omega .61 is below the review reference .70. Inspect
#> score purpose, indicator quality, dimensionality, and CFA evidence before
#> changing the scale.
#> - speed (Review): Omega .69 is below the review reference .70. Inspect score
#> purpose, indicator quality, dimensionality, and CFA evidence before
#> changing the scale.
#> - visual (Review): Alpha .63 is below the review reference .70. Inspect
#> score purpose, indicator quality, dimensionality, and CFA evidence before
#> changing the scale.
#> - speed (Review): Alpha .69 is below the review reference .70. Inspect score
#> purpose, indicator quality, dimensionality, and CFA evidence before
#> changing the scale.
#> - Review: At least one CFA fit reference is flagged for review. Model-based
#> omega is conditional on the adequacy of the fitted measurement model.
#> Investigate the CFA before treating the reliability coefficient as stable
#> evidence.
#>
#> Sampling uncertainty was not bootstrapped. For report-ready intervals, rerun
#> with `ci = "bootstrap"`.
#> Omega is primary for the congeneric CFA workflow; alpha is secondary and
#> assumption-dependent. Reliability contributes score-precision evidence, not
#> construct validity.
#>
#> See x$decision_log for the reasoning behind each coefficient.
rel$decision_log
#> # A tibble: 9 × 10
#> stage object metric value reference severity observation recommendation
#> <chr> <chr> <chr> <dbl> <chr> <chr> <chr> <chr>
#> 1 reliability measu… coeff… NA Dunn et … info Model-base… Interpret rel…
#> 2 reliability measu… model… NA Bell, Ch… review At least o… Investigate t…
#> 3 reliability visua… alpha… NA Dunn et … info Coefficien… Report alpha …
#> 4 reliability visual omega 0.612 configur… review Omega (.61… Inspect score…
#> 5 reliability textu… omega 0.885 configur… info Omega (.89… Carry this re…
#> 6 reliability speed omega 0.686 configur… review Omega (.69… Inspect score…
#> 7 reliability visual alpha 0.626 configur… review Alpha (.63… Inspect score…
#> 8 reliability textu… alpha 0.883 configur… info Alpha (.88… Carry this re…
#> 9 reliability speed alpha 0.688 configur… review Alpha (.69… Inspect score…
#> # ℹ 2 more variables: decision <chr>, rationale <chr>
# \donttest{
# Bootstrap intervals refit the same CFA repeatedly; use more resamples
# (for example 1000) for final reporting.
rel_ci <- nomo_reliability(cfa, ci = "bootstrap", ci_boot = 100, ci_seed = 2026)
summary(rel_ci)
#> <nomo_reliability summary> Reliability
#> Constructs: 3 | Review reference: .70
#>
#> Coefficients
#> Construct Indicators Omega 95% CI Alpha 95% CI Flag
#> visual continuous .61 [.54, .69] .63 [.55, .69] Review
#> textual continuous .89 [.86, .90] .88 [.85, .90]
#> speed continuous .69 [.59, .75] .69 [.63, .74] Review
#>
#> Flagged
#> - visual (Review): Omega .61 is below the review reference .70. Inspect
#> score purpose, indicator quality, dimensionality, and CFA evidence before
#> changing the scale.
#> - speed (Review): Omega .69 is below the review reference .70. Inspect score
#> purpose, indicator quality, dimensionality, and CFA evidence before
#> changing the scale.
#> - visual (Review): Alpha .63 is below the review reference .70. Inspect
#> score purpose, indicator quality, dimensionality, and CFA evidence before
#> changing the scale.
#> - speed (Review): Alpha .69 is below the review reference .70. Inspect score
#> purpose, indicator quality, dimensionality, and CFA evidence before
#> changing the scale.
#> - Review: At least one CFA fit reference is flagged for review. Model-based
#> omega is conditional on the adequacy of the fitted measurement model.
#> Investigate the CFA before treating the reliability coefficient as stable
#> evidence.
#>
#> What these columns mean
#> CI -- Percentile bootstrap confidence interval.
#>
#> Omega is primary for the congeneric CFA workflow; alpha is secondary and
#> assumption-dependent. Reliability contributes score-precision evidence, not
#> construct validity.
#>
#> See x$decision_log for the reasoning behind each coefficient.
# }