Journal of Productivity Analysis, 13, 249–262 (2000) c  2000 Kluwer Academic Publishers, Boston. Manufactured in The Netherlands. Efﬁciency Measurement With Unbalanced Panel Data: Evidence from Tunisian Textile, Clothing and Leather Industries MOHAMED GOA ¨ IED m.goaied@planet.tn LEA, FSEGT, University of Tunis III, 1060 Tunis, Tunisia RYM BEN AYED-MOUELHI rim.mouelhi@iscae.rnu.tn LEA, ISCAE, University of Tunis III, Rue des Entrepreneurs, 2035, La Charguia II, Tunis-Carthage, Tunisia Abstract This paper is concerned with the estimation of stochastic frontier production functions with unbalanced panel data when unobservable ﬁrm efﬁciency levels are related to explanatory variables. We use the ≪weighted-means≫ Instrumental Variables method acknowledged to R. Gardner (1998) which is a modiﬁcation of the Hausman-Taylor (1981) procedure adapted for unbalanced panel data. The estimation method is used to examine technical efﬁciency in Tunisian textile, clothing and leather (TCL) industries during the period 1983– 1994. Further, we assume a Translog production frontier where input uses are expressed in efﬁciency units and adjusted for the age of capital and types of labor. Firm-speciﬁc time-invariant technical efﬁciency is obtained using Schmidt and Sickles (1984) approach. The results suggest that the Instrumental Variables method produces more accurate estimates of the unknown ﬁrm level technical efﬁcency. Mean efﬁcency scores resulting from the MHT method is of 66.5%. Keywords: stochastic frontiers, technical efﬁciency, unbalanced panel data, instrumental variables, textile indus- try, Tunisia 1. Introduction Panel data offers two signiﬁcant potential advantages in the estimation of efﬁciency frontiers. 1 First, estimation of ﬁrm-speciﬁc technical efﬁciency may be performed without employing a strong distributional assumption of the errors. Second, it no longer calls for the need to assume that technical efﬁciency is independent of the inputs but still allows it to be tested avoiding the drawbacks of ﬁxed effects (Schmidt and Sickles (1984), Hallan and Machado (1995)). A correlation of inefﬁciency with regressors will yield biased and inconsistent estimates of parameters in OLS and GLS models. A common method for handling this biasedness is to use the ﬁxed effects estimation procedure. This within-groups estimation is not fully efﬁcient due to lost information contained in the eliminated means. Parameter estimation of the time invariant variables is impossible as these variables are eliminated in the demeaning