Silico by AgEcon | Silicon samples for choice experiments, experimental auctions and behavioural models

Silico by AgEcon

Silicon samples for choice experiments, experimental auctions and behavioural models

Before spending money on fieldwork, run your questionnaire on synthetic respondents: a language model answers your choice tasks or places bids as people with the profiles you specify. You get the data in the format you would analyse, so you can check the design, the wording and the analysis code in an afternoon.

Choice experiments

attributes and levels, 2–4 alternatives, opt-out, blocks
paste an efficient design from Ngene, Stata or R, or generate a random level-balanced one
long-format data for cmclogit, cmxtmixlogit, apollo, mlogit

Experimental auctions

BDM (random price) or Vickrey (second price)
several products, randomised order
information treatments between subjects, or within subjects (the same respondents bid again after the information)

Behavioural models

ready-made models: items generated from your behaviour, technology or risk, then editable
or your own attitudinal scales (Likert, semantic differentials, reverse-coded items)
wide-format data; do-file with reliability, factor analysis, regressions and SEM

Respondents

profiles drawn from the margins of your target population (gender, age, education, income, area…)
reproducible from a seed
Claude Haiku or Sonnet, temperature of your choice

What you get

data ready for analysis (CSV, long format)
raw answers, with the reason for every invalid one
a codebook with model, temperature, seed, dates and full prompts
a Stata file (.dta) with variable labels, and a do-file written for your design

Behavioural models available

Model	Constructs	Typical use	Source
Theory of Planned Behaviour (TPB)	attitude, subjective norm, perceived behavioural control, intention	food choices, sustainable consumption, farmers' decisions	Ajzen (1991)
Extended TPB	TPB + moral norm, past behaviour	ethical and pro-environmental behaviours	Ajzen (1991); Conner & Armitage (1998)
Technology Acceptance Model (TAM)	perceived usefulness, perceived ease of use, attitude toward using, behavioural intention	adoption of digital and precision agriculture tools	Davis (1989)
UTAUT2	performance and effort expectancy, social influence, facilitating conditions, hedonic motivation, price value, habit, behavioural intention	consumer acceptance of technologies, apps, novel foods	Venkatesh, Thong & Xu (2012)
Value-Belief-Norm theory (VBN)	biospheric values, awareness of consequences, ascription of responsibility, personal norm, intention	pro-environmental behaviour, sustainable diets	Stern et al. (1999); Stern (2000)
Protection Motivation Theory (PMT)	perceived severity, vulnerability, response efficacy, self-efficacy, response costs, protection motivation	climate adaptation, risk management, food safety	Rogers (1975, 1983)
Free scales	your own sections and items	any attitudinal battery	—

Items are generated from the construct definitions and the authors' guidelines as a starting point: for a publication, use the validated wording of the original source and cite it. Respondents never see construct names or item codes.

How it works

Design the experiment in the form: scenario, attributes or products, respondent profiles. Nothing is sent to the model.
Check the prompts each respondent will receive.
Pretest it on 3 respondents, immediately and with no authorisation: see the answers and download the data, the Stata file and the do-file, to check that everything works.
Request authorisation for the whole study. API calls are paid by the Observatory, so each study is reviewed by the administrator. Once authorised, launch it: results usually arrive within an hour.
Download the data, the raw answers and the codebook.

A pre-testing tool, not a sample. Synthetic respondents help to check that tasks are understood, to spot dominant levels, to obtain priors for an efficient design and to teach. They are not a substitute for human respondents: silicon samples typically show lower variance, stereotyped profiles and sensitivity to wording, and reflect the model's training data. Whenever you use them, report model, temperature, seed, dates and prompts.

Who can use it

Researchers with a verified profile in the Observatory. Sign in to design a study.