Package: ackwards 0.2.0.9000

The three engines at a glance | Setup | PCA: components from total variance | EFA: factors from common variance | How close are EFA and PCA loadings? | ESEM: EFA with full model diagnostics | Loading standard errors and confidence intervals | Per-level fit: what it tells you (and what it does not) | The key distinction | Reporting fit with tidy() and autoplot() | Should you care about fit in a bass-ackwards workflow? | How much do the edges differ? | Choosing an engine | Orthogonal or oblique rotation | Varimax and oblimin on the same data | How to decide | Missing data | FIML for continuous PCA/EFA | Which option to use? | Correlation-matrix input | Constraints | suggest_k() with a correlation matrix | Performance with many items (ESEM) | References

The problem: which solution is right? | Data | Step 1: How many factors? suggest_k() | Step 2: Fit the hierarchy ackwards() | Step 3: Summarize the result | High-level summary | Step 4: Visualize the hierarchy autoplot() | Adjusting the diagram | Step 5: Dig into the factors tidy() | Reading each factor with top_items() | Factor loadings (tidy) | Between-level edges | Variance explained | Step 6: Score observations augment() | Using scores for downstream analysis | Scoring new data (train/test) | Summary: the basic toolkit | Next steps | References

Why Pearson underestimates for ordinal items | The polychoric solution | Automatic detection | Seeing the difference | Correlation matrices | Loadings | Between-level edges | WLSMV for ESEM with ordinal items | Practical guidance: when to use polychoric | References

The problem: depth is where bass-ackwards analyses go wrong | Three verbs, three different questions | Setup | Step 1: Screen the plausible depth range with suggest_k() | Step 2: Gate the depth on replicability with comparability() | Step 3: Fit the hierarchy to the gated depth | Step 4: Find what perpetuates without differentiating, with prune() | How sure are we about these edges? boot_edges() | Step 5: Interpret with the built-in guardrails | Step 6: Validate downstream, out of sample | Common mistakes this workflow prevents | Summary: the workflow in six calls | References

The limitation of adjacent-only edges | Setup | Computing every between-level correlation with pairs = "all" | Reading the skip-level edge table | Reading the strongest edge with care | Pruning: identifying redundant factors | Chains of near-identical factors with prune(x, "redundant") | The pruned-factor diagram | Inspecting structural similarity with prune(x, "artifact") | The near-redundant band | Structural artifact signals | Tuning the thresholds | Practical interpretation | References

k is a decision, not an output | The five criteria | PA-PC: parallel analysis on the PC basis | PA-FA: parallel analysis on the FA basis | MAP: minimum average partial | VSS-1 and VSS-2: very simple structure | CD: comparison data | Summary table | Running suggest_k() | The diagnostic plot | Matching the criterion to your engine | The arguments | criteria: which criteria to compute | cor: correlation basis | n_iter: Monte Carlo iterations | seed: reproducibility | k_max: ceiling for the search | When the criteria agree, and when they do not | A worked recommendation for the BFI | References

Setup | Encoding sign and magnitude | sign_by: how direction is shown | magnitude_by: how |r| is shown | Filtering edges | cut_show: minimum |r| to display | Edge colors | color_pos / color_neg: custom direction colors | Monochrome mode | mono = TRUE: a black-and-white convenience wrapper | Correlation labels | show_r / r_digits: annotate edges with r values | Node labels | node_labels: rename individual factors | node_width / node_height: size individual boxes to fit a label | label_template(): generate the scaffold | Item content | show_items: list the items under the deepest-level factors | Structural simplifications | primary_only = TRUE: show only primary-parent edges | order: arrange the deepest level by hand | drop_pruned + show_secondary: the pruned view and its hidden correlations | Level labels | show_level_labels / level_label_size | Arrowheads | show_arrows = FALSE: plain line ends | Edge width | edge_linewidth: uniform vs. |r|-scaled width | Layout orientation | direction = "horizontal": left-to-right layout | Legend | legend = FALSE: suppress all guides | Worked example: publication-ready figure | Saving plots | Diagnostic scree / criteria plot: autoplot.suggest_k() | References

The data | Fitting the 10-level hierarchy | Skip-level structure | The redundancy chase | A chain, worked | Why the direct criterion matters here | The pruned-factor diagram | Fidelity | References

Reading a factor with top_items() | Cross-loadings are signal | Showing item wording instead of codes | The sign convention: negative does not mean "low" | Naming in a hierarchy | Borrowing names for the upper levels | Applying your names | Making the names stick with set_factor_labels() | The Forbes letter convention | Starting from a blank slate | References