BGPT: Paper Review: How to Model in Zoology: Basic Concepts and Explanations

Explore by Goal

Quick Answer Copied

Quick take:

The paper is a very short conceptual primer on using AIC, BIC, and DIC for model comparison—mostly asserting intended interpretations (quality/complexity/likelihood-like fit) without showing derivations, definitions, or assumptions.
AIC and BIC are real and well-established criteria, but the text’s mapping (“BIC measures complexity”, “DIC like r²”) is oversimplified and sometimes incorrect or at least non-precise for how DIC is actually used.

Long Answer

Paper Review (Evidence-First): “How to Model in Zoology: Basic Concepts and Explanations”

DOI: 10.23880/izab-16000556 • Published: 2024-02-07

What the paper claims (as written)

It presents AIC, BIC, and DIC as tools to compare statistical models and frames them as trading off quality, complexity, and reliability/likelihood-like accuracy.
It states that lower AIC means higher quality and suggests AIC is the best criterion in general.
It characterizes BIC as an abstract/complexity measure and claims higher BIC corresponds to being “more abstract” and “less pragmatic”.
It claims DIC is “similar to r²” and that higher DIC means “more value” / “more explanatory power”.

These claims are directly from the paper text.

Known foundations (what the classic sources actually establish)

AIC was introduced as a method for statistical model identification/selection based on an information-theoretic criterion.
BIC (Schwarz’s criterion) is derived from approximation ideas and is commonly presented as penalizing model complexity (with penalty tied to sample size under assumptions).
DIC is associated with Bayesian model comparison/estimation of deviance in hierarchical Bayesian settings; it is not generally “r²”.

Core conceptual structure (visual)

Note: the bar heights here are not from the paper as numbers; they only encode which interpretation the paper emphasizes (quality/complexity/likelihood-like). The actual AIC/BIC/DIC formulas and scaling are not provided in the article.

Critical appraisal (skeptical, evidence-based)

1) Missing definitions & assumptions

The paper does not provide the explicit AIC/BIC/DIC formulas, the assumptions under which their interpretations hold, or the estimation context needed for DIC (hierarchical Bayesian deviance framework).
It also does not explain when “lower is better” applies to the different criteria; it only states that lower AIC is good, and it gives an unusual “higher DIC is better” framing.
Because these details are absent, readers cannot verify whether the described “meaning” matches the criteria used in real modeling workflows.

2) Oversimplified or potentially misleading interpretation

The paper’s r² analogy for DIC is a strong interpretation leap. DIC is a deviance-based Bayesian model comparison criterion; mapping it to a frequentist coefficient of determination is not faithful in general.
It claims “the best criteria index… is the AIC” without stating the criterion-specific regimes where different choices can be appropriate (e.g., predictive performance vs approximation goals).

3) Zoology-specific guidance is essentially absent

Despite the zoology framing, the manuscript provides no zoology modeling worked examples (e.g., observation model vs process model separation, phylogenetic structure, detection bias), no data-collection linkage, and no discussion of what “scenarios” mean in ecological inference.
As a result, the paper functions more as a generic statistics mini-FAQ than as zoology-focused modeling instruction.

Reproducibility & usability

No empirical data, no modeling results, and no code are described; therefore there is nothing to reproduce beyond conceptual statements.
The reference list includes classical criterion origins (Akaike and Schwarz) and a DIC reference via Bayesian modeling documentation.
However, the article does not show how to compute AIC/BIC/DIC for a given zoological dataset or how to interpret them under common modeling structures.

What would most improve this paper (actionable checklist)

Add exact formulas for AIC and BIC (and specify whether small-sample correction AICc is intended).
For DIC, specify the Bayesian deviance framework, what “deviance” is computed from, and how DIC is expected to be interpreted in that context (including sign/directionality).
Explain that “best” depends on goal: predictive accuracy vs approximate posterior/complexity trade-offs, not a one-size-fits-all ranking.
Provide at least a single fully-worked zoology modeling example (process + observation, or detection/measurement models) showing how the criterion choice changes decisions.

Next-step links (BGPT)

Author reviews (BGPT)

Note: author-name parsing can be ambiguous in the provided TEI metadata, so if BGPT’s author pages don’t match, re-run with the exact author spelling shown in the publisher metadata.

Feedback:

Updated: April 17, 2026