Statistical Linear Regression Method Research Articles

Fetal lung volumetry using two‐ and three‐dimensional ultrasound

To compare methods of measuring fetal pulmonary volume and to establish nomograms of fetal pulmonary volume according to gestational age for the accurate diagnosis of pulmonary hypoplasia. Three methods of measuring fetal pulmonary volume in 39 normal fetuses were compared: two-dimensional (2D) ultrasound measurement assuming that the lung is a geometrical pyramid, three-dimensional (3D) ultrasound using the VOCAL rotational method, and the conventional multiplanar 3D mode. Linear regression was used to construct an equation for 3D volume calculation from 2D measurements (the re-evaluated pulmonary volume equation (RPVE)). Lung volume measurements were recorded from 622 singleton fetuses in order to construct nomograms. There was no statistically significant difference between the lung volume values obtained using the two 3D modes. However, in comparison with the 2D measurements the volumes obtained were larger (mean difference = 11.99, P < 0.1 x 10(-6)). The relationship between the 2D and 3D volumes was determined using a statistical linear regression method: RPVE (mL) = 4.24 + (1.53 x 2DGPV), where 2DGPV (2D geometric pulmonary volume) = (surface area right lung base (cm2) + surface area left lung base (cm2)) x 1/3 height right lung (cm). Two nomograms were constructed, one for use with 2D and one for 3D technology. 2D pulmonary volume assessment can be used in clinical situations where fetal prognosis depends on lung volume and its growth potential. It is routinely available and easy to perform particularly when repeat measurements are required in evaluation of lung growth. We therefore propose this method as an alternative to magnetic resonance imaging or 3D ultrasound.

A SAS macro for the analysis of cell survival curves

This paper describes a SAS macro for the statistical analyses of cell survival data obtained after radiation treatment using the methods of R.E. Tarone et al. (Mutation Research 111 (1983) 79–96). These analyses are usually required on a routine basis by all biomedical research laboratories involved in cell survival assays generating dose–response curves aimed at characterizing radiosensitive mutant cell strains or individuals whose body cells exhibit enhanced sensitivity to radiation and other genotoxic agents. Statistical methods of linear regression are applied to data from repeated experiments with a cell line/strain and weighted estimates of a common slope and its variance are obtained. The methods are currently implemented in two APL programs. These programs are not easily accessible to most biomedical statisticians and researchers because APL is not a common software tool for statistical analysis. Implementation of these methods in SAS, a widely used commercial software for statistical analysis, is expected to help resolve this issue. We illustrate the application of the macro using an example data set obtained in our laboratory, and hope that other investigators may find it useful in analyzing their data.

A comparison of linear regression and neural network methods for predicting excess returns on large stocks

Recent studies have shown that there is predictable variation in returns of financial assets over time. We investigate whether the predictive power of the economic and financial variables employed in the above studies can be enhanced if the statistical method of linear regression is replaced by feedforward neural networks with backpropagation of error. A shortcoming of backpropagation networks is that too many free parameters allow the neural network to fit the training data arbitrarily closely resulting in an "overfitted" network. Overfitted networks have poor generalization capabilities. We explore two methods that attempt to overcome this shortcoming by reducing the complexity of the network. The results of our experiments confirm that an "overfitted" network, while making better predictions for within-sample data, makes poor predictions for out-of-sample data. The methods for reducing the complexity of the network, explored in this paper, clearly help improve out-of-sample forecasts. We show that one cannot say that the linear regression forecasts are conditionally efficient with respect to the neural networks forecasts with any degree of confidence. However, one can say that the neural networks forecasts are conditionally efficient with respect to the linear regression forecasts with some confidence.