Who nomologR is for
nomologR is written for three audiences:
- graduate students (master’s and doctoral) who are learning scale development and construct validation and need to understand why each analysis is performed, not only how to run it;
- faculty and advisors who teach measurement and want output that explains its own reasoning; and
- researchers who need a defensible, reproducible record of how a measure was evaluated.
The package coordinates established engines — psych,
EFAtools, lavaan, and semTools —
and adds a teaching layer: diagnostics framed as evidence, explanations
tied to the methodological literature, and a record of every
consequential researcher decision.
The question behind every stage
At each step, nomologR tries to help answer one
question:
What does this result tell me about the claim that these observations measure the construct I say they measure, and what should I investigate next?
The package flags, explains, and documents. It never silently deletes items, respecifies models, or declares a scale “valid.”
What nomologR assumes about the items
Every analysis in nomologR assumes reflective
measurement: the construct causes the responses, so its items are
interchangeable manifestations of it and should correlate with one
another. That assumption is what makes factor loadings, internal
consistency, AVE, HTMT, and measurement invariance meaningful, and what
makes a weakly loading item worth reviewing (Bollen & Lennox,
1991).
Some measures are built the other way. In a formative
measure, the indicators define the construct, as income, education, and
occupation define socioeconomic status. They need not correlate, and
dropping one changes what is measured (Diamantopoulos & Winklhofer,
2001). The tools here do not apply to such a measure, and applying them
misleads: low internal consistency is expected, and deleting a “weak”
indicator removes part of the construct. Decide which model a measure
follows before choosing tools; Jarvis et al. (2003) and MacKenzie et
al. (2005) give criteria. Whether formative latent variables are
meaningful at all is still debated (Howell et al., 2007; Edwards, 2011;
Bollen & Diamantopoulos, 2017), and nomologR does not
fit them. A formative index can still enter a nomological network as an
observed variable in nomo_network(), and its content can
still be reviewed with contentvalidR.
The workflow at a glance
| Stage | Question it answers | Functions | Walkthrough |
|---|---|---|---|
| Planning the sample | How many cases does the planned model need? |
nomo_power_rmsea(),
nomo_power_simulate()
|
From CFA to a defensible measurement model |
| 1. Item and data audit | What do the items and responses look like? | nomo_screen() |
From item audit to exploratory structure |
| 2. Dimensionality | How many latent dimensions deserve investigation? | nomo_factors() |
From item audit to exploratory structure |
| 3. Exploratory structure | What does a requested factor solution look like? | nomo_efa() |
From item audit to exploratory structure |
| 4. Confirmatory measurement model | Does a prespecified model reproduce the data, and where is there strain, such as cross-loadings or shared method? |
nomo_model(), nomo_split(),
nomo_cfa(), nomo_esem(),
nomo_method_variance()
|
From CFA to a defensible measurement model |
| 5. Reliability and construct-validity evidence | How precise and how stable are scores, and are constructs distinguishable? |
nomo_reliability(), nomo_retest(),
nomo_validity()
|
From CFA to a defensible measurement model |
| 6. Generalizability | Is the construct measured comparably across groups and occasions, and do their latent means differ? |
nomo_invariance(),
nomo_invariance_longitudinal(),
nomo_partial()
|
Measurement invariance |
| 7. Nomological network | Does the construct relate to others as theory predicted? |
nomo_hypotheses(), nomo_network(),
nomo_single_indicator()
|
Theory-specified nomological networks |
| Guided workflow | How do the stages fit together with explicit decisions? | nomo_run() |
Guided workflow |
| Starting from content review | What does empirical evidence add to an expert review of the items? |
nomo_screen() and nomo_run() with a
contentvalidR handoff |
From content review to empirical screening |
| Reporting | How do I archive the evidence and decisions? |
nomo_report(), nomo_table()
|
Archiving a workflow |
Teaching datasets
The examples and walkthroughs use simulated datasets whose population models are documented. Because the truth is known, you can check what the package reports against what actually generated the data — something real data never allow.
| Dataset | Design | What it teaches |
|---|---|---|
nomo_demo_continuous |
Two correlated factors, five candidate items each, 500 cases | A cross-loading item (a5), a weak item
(b5), and a small amount of missing data |
nomo_demo_ordinal |
The same latent responses cut into five ordered categories | Polychoric correlations, WLSMV estimation, and ordinal reliability |
nomo_demo_network |
Three constructs, an observed outcome, two administration groups, 800 cases | Theory-specified predictions, equivalence regions, replication, and a known source of scalar non-invariance |
nomo_demo_walkthrough |
Twelve items an expert panel reviewed in contentvalidR,
two facets, two cohorts, 400 cases |
Items that pass content review and fail empirically, and the reverse; reverse-worded items; a cohort difference |
nomo_demo_longitudinal |
Four Wellbeing items answered by the same people on three occasions, 500 cases | Longitudinal invariance, item residuals correlated across occasions,
latent change, and a known intercept drift in w3
|
See ?nomo_demo_continuous,
?nomo_demo_ordinal, ?nomo_demo_network,
?nomo_demo_walkthrough, and
?nomo_demo_longitudinal for the full population models.
A first look
An item audit describes the data without changing it:
scr <- nomo_screen(nomo_demo_continuous)
scr
#> <nomo_screen> Item and data audit
#> Cases: 500 | Candidate items: 10
#> Items with missing responses: 2 | Constant: 0 | All missing: 0
#> Items in correlation diagnostics: 10 | Item-rest correlations: 10
#> Response concentration flags: 0 | Near-zero variance: 0
#> Decision log: 4 entries | Flagged: 1 review, 0 concern
#> No rows or items were removed or modified.
#>
#> See summary(x) for each item's review and nomo_table(x, "decision_log") for
#> every log entry.Factor-retention evidence triangulates several criteria rather than trusting a single rule:
fac <- nomo_factors(nomo_demo_continuous, seed = 2026)
fac
#> <nomo_factors> Factor-retention evidence
#> Cases: 500 (minimum pairwise N: 473) | Items: 10 | Correlation: Pearson
#> Criterion set: core | Methods run: 3 | Families: 2 | Not run: 1 | Flags: none
#> Parallel analysis (percentile rule): 2 | MAP: 2 (TR2), 2 (TR4) | KMO: .87
#> MAP = Velicer's minimum average partial criterion, original (TR2) and revised
#> (TR4); KMO = Kaiser-Meyer-Olkin measure of sampling adequacy.
#>
#> Both available criterion families (3 methods) point to 2 factors. Agreement
#> between two criterion families is limited evidence for investigating that
#> solution, not proof of dimensionality. 1 requested method was not evaluated;
#> nomo_table(x, "criteria") gives the reason.
#>
#> See summary(x) for the evidence by method and the criteria that did not run.The population model for these data has two factors, and the retention evidence points to investigating two. Notice the wording: the evidence supports investigating a two-factor solution; it does not prove that the scale “has” two factors.
How to read nomologR output
- Observation, reason, options, consequence. Recommendations state what was found, why it may matter, what defensible choices exist, and what each choice changes downstream.
-
Reference values are prompts, not verdicts.
Familiar numbers such as a loading of 0.40 or a CFI of .95 flag evidence
for review. An item can be flagged for review or concern, never for
deletion. The references live in
nomo_defaults(), and changing them is a visible researcher choice. -
One flag wording on screen. Printed output, plots,
and reports use one flag wording: “review”, then “concern”, “not
computed” for a value that could not be computed, and no flag when no
reference fired (a plot legend calls that “No flag”). Plots show each by
shape as well as color. The returned tables keep each analysis’s own
values (
KEEP,REVIEW, andSTRONG REVIEWfor loadings;info,review, andconcernfor evidence), so code that filters on them is unaffected.?nomologRsets out how these words correspond to the statusescontentvalidRreports. - Historical methods are labeled. Techniques you will meet in published work — eigenvalues greater than one, coefficient alpha, the Fornell–Larcker comparison — are shown as context where useful, qualified, and kept separate from contemporary evidence.
-
Engine results are never hidden. Every result keeps
the underlying
psychorlavaanobject (usually in$fit) for direct inspection. -
Decisions are recorded. Each result carries a
decision log, and
nomo_run()records the researcher’s decisions and rationales across stages.
A suggested learning path
- From item audit to exploratory structure — screening, factor retention, and EFA.
- From CFA to a defensible measurement model — confirmatory fit, reliability, and convergent and discriminant evidence; then ESEM beside the CFA, test-retest reliability, method variance from a marker variable, and sample-size planning.
- Measurement invariance — comparability across groups, localized strain, researcher-controlled partial invariance, latent means as known-groups evidence, and invariance across occasions.
- Theory-specified nomological networks — predictions, equivalence regions, replication, and observed composites corrected for their unreliability.
- Guided workflow — the stages combined with explicit decisions.
-
From content review to empirical
screening — the items a
contentvalidRreview carried forward, screened, including where the two stages disagree. - Archiving a workflow — a reproducible report.
Throughout, the research basis
article explains where each method came from, what the contemporary
literature recommends, and what nomologR implements. Every
function’s help page lists its references.
Where nomologR is heading
The next release, v1.0.0, is planned together with contentvalidR,
which covers the content-review stage before this package’s. At 1.0.0
both packages’ interfaces become stable, and the handoff between them
becomes a supported contract. Larger extensions, such as IRT, multiple
imputation, and formative models, are candidates for later 1.x releases.
See the roadmap
and the v1.0.0
milestone.
Citing nomologR
citation("nomologR")
#> To cite nomologR in publications, please use:
#>
#> Uhalt, J. (2026). nomologR: Guided scale development and construct
#> validation (Version 0.9.0) [Computer software].
#> https://github.com/JUhalt/nomologR
#>
#> A BibTeX entry for LaTeX users is
#>
#> @Manual{,
#> title = {{nomologR}: Guided Scale Development and Construct Validation},
#> author = {Joshua Uhalt},
#> year = {2026},
#> note = {(Version 0.9.0) [Computer software]},
#> url = {https://github.com/JUhalt/nomologR},
#> }