Experimental Evidence on Combining Cross-Section and Time Series Information

摘要: IN recent years several research studies have used a data base consisting of time series cross-section samples. A primary reason is that panel this type are potentially richer in information than single sample. To date, however, the question how to best analyze bases has not been fully explored any research. illustrate, number prior projects which utilized briefly reviewed. Hoch (1962) moving samples as for estimating parameters CobbDouglas production function by analysis covariance. Specifically were collected on 63 Minnesota farms 1946 1951. reported an observed difference between least squares parameter estimates and covariance elasticities developed from these estimates. In terms method, major conclusion was model might produce less biased marginal return Massy Frank (1965) investigated relationship price changes dealing activities -firm's market share frequently purchased household food products. Panel covering 101-week period family purchase history provided study. However, aggregated so no methodological insight could be inferred concerning analyzing data. Laughhunn Lyon (1971) applied Bayesian regression cigarette consumption using Tiao Zellner (1964) approximation method. The issue observe differences exist classical pooling technique. Comparison two techniques revealed very little either or standard errors. Schipper (19541957) Survey Research Center, University Michigan. central focus study consumer discretionary behavior particularly with respect durable expenditures, short term debt, saving. finding individual regressions existed. suggested analyses but prove point since he did use experimental Palda Blair (1970) conducted toothpaste demand multiple cross sections MRCA during 1958-1962. One their investigate potential specification bias, based rationale presented Simon Aigner (1970), can because omitted variables. An interpretation results led them think may reduce bias. This cannot considered conclusive framework. Since there interest part economic business researchers sample data, would appear sufficient evaluating different methods available combining Earlier work area includes Nerlove (1967, 1968). He assumed models form Yit = aYit-l + Uit 1-Xit respectively yi Vit uncorrelated where =o-2 2 +or2. estimation OLS, generalized utilizing known p (p o-A2/o-X2) analysis-of-covariance cross-sectional effects only, two-round estimated value p, maximum likelihood Generally, Nerlove's findings indicated (if known) produces good esti-