The reliability of a composite modeled as a single indicator
Source:R/nomo_single_indicator.R
nomo_single_indicator.Rdnomo_single_indicator() records the reliability of an observed composite,
such as a scale mean or sum, so that nomo_network() can model the
composite as a single-indicator latent variable. The composite's error
variance is fixed at \((1 - \rho)\sigma^2\), where \(\rho\) is its
reliability and \(\sigma^2\) its variance as the model analyzes it, so the
relationships it enters are corrected for its unreliability. The variance
is computed on the cases the model analyzes (the complete cases, under
listwise deletion) and with the estimator's denominator, so the fitted
model's reliability for the composite is the one supplied.
Usage
nomo_single_indicator(
reliability,
se = NULL,
coefficient = c("omega", "alpha", "other"),
source = NULL,
construct = NULL
)Arguments
- reliability
The composite's reliability: one number strictly between 0 and 1, or a
nomo_reliability()result, whose omega forconstructis used.- se
Optional standard error of the reliability. With a
nomo_reliability()result that has bootstrap intervals, the default is approximated from the omega interval, as its width divided by twice the normal quantile of its level.- coefficient
Which coefficient
reliabilityis:"omega"(default),"alpha", or"other".- source
Optional text naming where the reliability comes from, such as a citation or "this sample". It is recorded in the decision log.
- construct
With a
nomo_reliability()result, the construct whose omega is used.
Value
A nomo_single_indicator object. The fields to read are
reliability, se (NA when not supplied), coefficient, source
(NA when not supplied), and construct (NA unless taken from a
nomo_reliability() result). print() shows the reliability, its
standard error, and where it comes from.
Details
Where the method comes from. Spearman (1904) corrected a correlation for the unreliability of its two measures. Structural equation modeling carried the idea into the model itself: a composite becomes the one indicator of a latent variable whose error variance is fixed from the composite's reliability, as the SEM textbooks describe (Hayduk, 1987; Bollen, 1989). Applied work in organizational research and marketing took it up where modeling every item was impractical (Williams & Hazer, 1986), and Bagozzi and Heatherton (1994) called it the "total aggregation" model.
Does it work? Savalei (2019) compared single indicators with path
analysis, which ignores measurement error, and with full multiple-indicator
SEM, in samples of 30 to 200. Path analysis and single indicators whose
reliability was fixed a priori, at a value that slightly overestimated the
true one, gave the most accurate estimates and the most power. Single
indicators whose reliability was estimated from the same data (coefficient
alpha) and full SEM performed in between, and she recommended a
fixed-reliability single indicator in small samples. An omega estimated from
the same sample, as from nomo_reliability() on these data, corresponds to
the data-estimated variant. Where the items are available and the sample is
large enough, modeling them as indicators remains the fuller correction.
Which reliability. A coefficient corrects only the error its design can
see: internal consistency, for example, leaves transient error in, and the
coefficient chosen decides what the correction means (DeShon, 1998). Omega
from the composite's own measurement model (nomo_reliability()) is the
default. Coefficient alpha understates reliability when loadings differ, so
it fixes too large an error variance and overcorrects; nomo_network() flags
it for review. The single indicator is the observed sum or mean, so with
ordered items the omega should be the observed-score one
(ordinal_scale = TRUE, the default); a latent-response omega
(ordinal_scale = FALSE) describes a hypothetical continuous composite, and
taking it brings a warning and is recorded in source.
Uncertainty. A reliability is itself an estimate. Standard errors that
treat it as known are too small, and Oberski and Satorra (2013) derived the
term to add. Supply se, and nomo_network() adds it. The derivation
treats the reliability as estimated independently of the data; when it comes
from the same sample, the added term is an approximation.
References
Bagozzi, R. P., & Heatherton, T. F. (1994). A general approach to representing multifaceted personality constructs: Application to state self-esteem. Structural Equation Modeling, 1(1), 35-67. doi:10.1080/10705519409539961
Bollen, K. A. (1989). Structural equations with latent variables. Wiley. doi:10.1002/9781118619179
DeShon, R. P. (1998). A cautionary note on measurement error corrections in structural equation models. Psychological Methods, 3(4), 412-423. doi:10.1037/1082-989X.3.4.412
Hayduk, L. A. (1987). Structural equation modeling with LISREL: Essentials and advances. Johns Hopkins University Press.
Oberski, D. L., & Satorra, A. (2013). Measurement error models with uncertainty about the error variance. Structural Equation Modeling, 20(3), 409-428. doi:10.1080/10705511.2013.797820
Savalei, V. (2019). A comparison of several approaches for controlling measurement error in small samples. Psychological Methods, 24(3), 352-370. doi:10.1037/met0000181
Spearman, C. (1904). The proof and measurement of association between two things. The American Journal of Psychology, 15(1), 72-101. doi:10.2307/1412159
Williams, L. J., & Hazer, J. T. (1986). Antecedents and consequences of satisfaction and commitment in turnover models: A reanalysis using latent variable structural equation methods. Journal of Applied Psychology, 71(2), 219-231. doi:10.1037/0021-9010.71.2.219
See also
nomo_network(), whose single_indicators argument takes these
records.
Examples
# A published reliability, with its source.
nomo_single_indicator(.85, source = "Test manual, Table 4")
#> <nomo_single_indicator> Reliability for a single indicator
#> Reliability: .850 (omega)
#> Source: Test manual, Table 4
#>
#> See ?nomo_network for how x in `single_indicators` fixes the composite's error
#> variance.
# Omega from the composite's own measurement model in this sample.
cfa <- nomo_cfa("Persistence =~ pe1 + pe2 + pe3 + pe4", nomo_demo_network)
nomo_single_indicator(nomo_reliability(cfa), construct = "Persistence")
#> <nomo_single_indicator> Reliability for a single indicator
#> Reliability: .816 (omega) | Construct: Persistence
#> Source: this sample (nomo_reliability())
#>
#> See ?nomo_network for how x in `single_indicators` fixes the composite's error
#> variance.