Tobit Censored Regression Model
Also known as: censored regression, limited dependent variable model, Tobit Modeli (Sansürlü Regresyon)
The Tobit model is a regression for outcomes that are censored at a threshold, estimating the relationship by maximum likelihood. Introduced by James Tobin in 1958, it addresses the pile-up of observations at a limit (typically zero) in data such as spending, wages, or duration.
Key highlights
- Correctly handles outcomes censored at a threshold, removing the bias that ordinary regression suffers when many observations stack at the limit.
- Maximum-likelihood estimation gives consistent coefficients that describe the underlying latent relationship.
- Naturally fits common economic data with mass at zero, such as spending, wages, or durations.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use the Tobit model when the continuous outcome is censored at a known limit — most often a lower bound of zero — and a noticeable share of observations pile up exactly at that limit, as with expenditure, wage, or duration data. It assumes a normally distributed error term and a known censoring point, and a large sample (at least about 100 observations) is recommended for the maximum-likelihood estimates to be reliable. It is appropriate for cross-sectional and panel structures aimed at prediction or explaining relationships.
Strengths & limitations
- Correctly handles outcomes censored at a threshold, removing the bias that ordinary regression suffers when many observations stack at the limit.
- Maximum-likelihood estimation gives consistent coefficients that describe the underlying latent relationship.
- Naturally fits common economic data with mass at zero, such as spending, wages, or durations.
- Relies on a normally distributed error term; departures from normality bias the estimates.
- The censoring point must be known in advance, typically fixed at zero.
- Maximum-likelihood consistency depends on a large sample, so small datasets are unreliable.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How does the Tobit model differ from ordinary regression?
Ordinary regression treats values at the censoring limit as ordinary observations, which biases the slope when many outcomes stack at the limit. The Tobit model models a latent outcome and adds the probability of being censored to the likelihood, so the estimates describe the underlying relationship rather than the truncated record.
What is the censoring point and does it have to be zero?
The censoring point is the threshold below (or above) which the outcome is not observed and is instead recorded at the limit. It must be known in advance; in practice it is most often a lower bound of zero, as with spending or duration data.
Why does the Tobit model need a large sample?
The coefficients are estimated by maximum likelihood, whose consistency and reliability rest on large-sample properties. A sample of at least about 100 observations is recommended; with too few observations the estimates become unstable.
When should I prefer a hurdle or two-part model instead?
If the zeros reflect a genuine decision not to participate rather than a censored positive value, a hurdle or two-part model that separates the participation decision from the amount may describe the data better than a single Tobit specification.
Sources
- 1.Tobin, J. (1958). Estimation of Relationships for Limited Dependent Variables. Econometrica, 26(1), 24-36.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 1). Tobit Model. ScholarGate. https://scholargate.app/econometrics/tobit-model