Skip to content
No Vape

Read AIC and BIC in a vaping-study model table

Keep relative model-fit comparisons separate from proven classes of people, a health effect and a certificate that the selected model is true.

No Vape · Updated

Books arranged on a shelf
Illustrative image · Pramod Tiwari / Unsplash

Find what each row changes

A vaping paper may compare different numbers of latent classes, predictor sets or time structures. Read the row labels and the actual criterion heading. A number in an AIC column is not a count of participants, the number of people who quit or a probability that a class describes you.

The original young-adult device-and-flavor study using 2018–2019 data presents a latent-class fit table with AIC, BIC and parameter counts among its columns. That is a concrete setting in which several candidate models receive different summaries. The presence of such a table does not establish an individual diagnosis or demonstrate a cessation benefit.

Original research / Addictive Behaviors: Young adult e-cigarette use: A latent class analysis of device and flavor use, 2018–2019

Keep fit and complexity in the same sentence

Stata’s official information-criteria manual describes these measures as functions of the likelihood, estimated parameter count and, for some criteria, sample size. It treats them as comparisons among models fit to the same dataset. A model can accommodate more detail while paying a larger complexity penalty; a label such as best therefore needs the named criterion.

In a fictional reading example, a two-pattern model uses fewer parameters while a three-pattern model fits observed answers more closely. You record both candidates and the criterion each comparison uses. Do not conclude that three permanent types of people exist simply because one candidate received a favorable score. This example has no calculated fit values or real vaping findings.

StataCorp: estat ic — Display information criteria

Allow criteria to point in different directions

AIC and BIC are different criteria, so they need not select the same candidate. Keep the authors’ stated choice and rationale visible rather than switching columns to obtain the conclusion you prefer. If the analysis changed the dataset or outcome between rows, that is a comparability question to resolve before accepting an ordinary score comparison.

The manual explicitly discusses criteria producing different preferences. That is not automatic proof of misconduct or a reason to declare the entire report useless. It means the selection rule and substantive purpose matter. More specialized fitting conditions can require methodological expertise; this guide does not tell an analyst which model to fit.

StataCorp: estat ic — Display information criteria

Finish with a relative claim

Your note can say: “Among these stated candidates on these data, this criterion favored this model; the authors also considered these other factors.” Keep classification uncertainty, external replication and evidence of any intervention benefit on separate lines. A relative fit comparison does not answer those questions by itself.

If the candidate set or selection rationale remains unclear, leave that uncertainty in the summary. Do not assign yourself to a research profile or choose treatment from a table of model criteria. The useful outcome is a faithful description of the comparison the paper actually made, without letting a technical best-fit label become a personal health conclusion or a promise about a tracking app.

Sources and scope

Sources inform the guidance. The worksheets and examples are No Vape editorial suggestions, not tested treatments.

  1. Original research / Addictive Behaviors — Young adult e-cigarette use: A latent class analysis of device and flavor use, 2018–2019
  2. StataCorp — estat ic — Display information criteria
  3. No Vape — No Vape — app features and store links

Written with AI assistance and checked against the cited sources. No clinical review is claimed. This guide covers practical planning, not individual medical treatment. How we research and write