Report R2.17 Quantitative Empirical Analysis in Management Research Seminar Leader: V. Nagi Reddy, IIM Calcutta (Retired) vnr[at]iimcal.ac.in Regression models can be broadly classified into linear and nonlinear types. This can further be classified into deterministic and stochastic models. At a basic level, one aims to find out the existence of a relationship between independent and dependent variables. Data could be time-series data or cross-sectional data. To draw a sample, we require a sampling frame. Convenient sample should reflect the target population that a person wants to study. Crosschecks need to be done to examine whether the data are collected in a random manner or not. The seminar leader opined that transparency in sampling would contribute to the objectivity of the study. He also suggested that, in general, systematic sampling yielded better results than most other sampling techniques. There are two variables that determine the size of the sample. One is the permissible error and the other is the level of confidence. In general, a smaller sample size would yield a larger error. As the rarity of the phenomenon studied increases, or the variability of data decreases, one requires larger sample sizes to obtain the same level of confidence. The seminar leader examined the basic multivariate regression model and assumptions in that model. The input variables have to be independent, the sample size should be at least two more than the number of independent variables, and the error in the model should be normally distributed with mean zero and variance Sigma Square. One of the most important assumptions in the multivariate regression model is that of homoscedasticity, which implies that, the variance is constant throughout the observations. The seminar leader elaborated on tests that need to be done to find out whether in a given data and model the error is homoscedastic or not. In case this assumption does not hold true, various methods for pre-processing the data have to be applied so that homoscedasticity is attained. Gauss Least Square Method is used to find the values of the coefficients of the model. This estimator is considered to be the Best Linear Unbiased Estimator (BLUE). The basics of time series modeling, including ARIMA models were briefly discussed. The possibility of serial correlation in time-series data was also highlighted. Reported by K. M. Rajesh and Jacob D. Vakkayil. Copyleft Xavier Institute of Management, Xavier Square, Bhubaneswar 751013, India Research World (ISSN 0974-2379) http://www1.ximb.ac.in/RW.nsf/pages/Home |
