Bias-Corrected Least Squares Dummy Variable (LSDVC) Estimator
Also known as: Bias-Corrected LSDV, BC-LSDV, Kiviet Estimator, Önyargı Düzeltilmiş En Küçük Kareler Kukla Değişken Tahmincisi
LSDVC is a bias-corrected panel data estimator introduced by Kiviet (1995) to address the well-known Nickell bias that afflicts the standard Least Squares Dummy Variable (LSDV) estimator in dynamic panel models with a lagged dependent variable. It is particularly suited for researchers working with datasets where the number of time periods T is small relative to the number of cross-sectional units N, such as firm-level or country-level panels spanning a short time horizon.
Key highlights
- Substantially reduces Nickell bias in short dynamic panels without discarding within-group variation
- More efficient than IV and GMM alternatives in small-to-moderate samples, often yielding lower root mean squared error
- Produces estimates that are asymptotically normal, enabling standard inference and hypothesis testing
- Does not require external instruments beyond those needed for the first-stage consistent estimator
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
LSDVC is most appropriate for balanced or nearly balanced dynamic panel datasets with a small number of time periods (T roughly 5–30) and a moderately large cross-section (N > T). The model must contain a lagged dependent variable and individual fixed effects. It outperforms GMM estimators in small samples because it exploits the efficient LSDV variance structure rather than relying on potentially weak instruments. It requires a consistent first-stage estimator and performs best under homoskedastic, serially uncorrelated errors. For very large T, the Nickell bias diminishes and standard LSDV suffices; for very large N with weak instruments, Arellano–Bond or Blundell–Bond GMM may be preferred.
Strengths & limitations
- Substantially reduces Nickell bias in short dynamic panels without discarding within-group variation
- More efficient than IV and GMM alternatives in small-to-moderate samples, often yielding lower root mean squared error
- Produces estimates that are asymptotically normal, enabling standard inference and hypothesis testing
- Does not require external instruments beyond those needed for the first-stage consistent estimator
- Requires a balanced (or near-balanced) panel; unbalanced structures complicate the analytical bias formula
- The bias approximation is derived under strict exogeneity of regressors other than the lagged dependent variable; predetermined or endogenous covariates violate this assumption
- Relies on the accuracy of the first-stage consistent estimator: if the first stage is poor (e.g., weak instruments), the bias correction itself is mis-estimated
- Asymptotic properties are established for large N with fixed T; the estimator is not designed for large-T panels where alternative approaches dominate
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 LSDVC differ from Arellano–Bond GMM?
Arellano–Bond GMM instruments the lagged dependent variable with further lags in first differences, which is consistent but can suffer from weak instrument problems and efficiency loss in small samples. LSDVC instead starts from the efficient LSDV estimator and analytically corrects its bias, typically achieving lower mean squared error when N is moderate and instruments are not particularly strong.
Which first-stage consistent estimator should I use with LSDVC?
Kiviet (1995) and Bruno (2005) discuss three choices: Anderson–Hsiao IV, Arellano–Bond GMM, and Blundell–Bond system GMM. In practice, Arellano–Bond is most commonly used as the initializer. The choice matters mainly for small N; as N grows, all three first-stage options yield similar bias corrections.
Can LSDVC handle time-varying covariates that are endogenous?
No. The analytical bias derivation in Kiviet (1995) assumes strict exogeneity of all regressors except the lagged dependent variable. If additional regressors are predetermined or endogenous, the bias formula is invalid, and GMM-based estimators that instrument those covariates are more appropriate.
Sources
- 1.Kiviet, J. F. (1995). On bias, inconsistency, and efficiency of various estimators in dynamic panel data models. Journal of Econometrics, 68(1), 53–78.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 2). LSDVC. ScholarGate. https://scholargate.app/econometrics/lsdvc