productivity. After an attempt to use the raw data, yield and height 

 were transformed to their logarithm (base 10) to provide a better fit for 

 the curves, decrease random variation, and normalize the mean square 

 error. There are 48 independent variables in the original data which 

 consisted of the concentration of 11 different mineral nutrients in the 

 plant tissue at each of 3 sampling times, the salinity of the soil 

 solution at each sampling time, and 12 soil chemical and physical proper- 

 ties. The "best" regression model was selected using a combination of 

 the maximum R 2 (multiple correlation coefficient) improvement procedure 

 (Service, 1972), the stepwise regression procedure (Draper and Smith, 

 1966), and critical examination of the independent variables from an 

 agronomic view. In the stepwise procedure, the single variable model 

 which produces the highest R 2 is selected; then variables are added one 

 by one to the model according to their significance measured by the 

 F-test (0.1 level of significance for entry). At every stage of regres- 

 sion, the variables incorporated in the previous stages are reexamined 

 for significance. A variable entered at an early stage may lose its 

 significance because of its relationship to other variables entering the 

 regression. The maximum R 2 improvement technique developed by J. H. 

 Goodnight (Department of Statistics, North Carolina State University, 

 Raleigh, North Carolina) selects the "best" one-variable model, the "best" 

 two-variable model, etc. according to R 2 . 



b. Selecting the Dependent Variable . The independent variables were 

 measurements of soil chemical and physical properties and nutrient con- 

 centrations in the plant tissue at three sampling dates (Tab. 27). 

 Multiple regression procedures were used as variable screening devices to 

 determine which of these independent variables were related to yield and 

 height of S. alt erni flora. It was necessary to select two models (one 

 for Iog io y ield and one for lo §10 height), because the relationship 

 between height and yield was not as close as might be expected (Fig. 55) . 

 Although the relationship between yield and height is highly significant, 

 the R 2 is only 0.26. The R 2 was not greatly improved by transforma- 

 tion to logarithms. Stands are thinnest where the tallest grass occurs; 

 consequently, shorter grass may produce higher yields where stands are 

 thicker. An example is seen by comparing the data for yield and height 

 from the tall height zones at Oregon Inlet and Beaufort (Tab. 28). The 

 average height at Beaufort is much greater than at Oregon Inlet, but the 

 yield at Oregon Inlet is greater. This is attributed to there being 

 nearly twice as many stems per unit area at Oregon Inlet to produce the 

 total biomass. This difference in growth habit may be related to tide 

 range or possibly to genetic difference between the grass from different 

 locations. Shorter grass and thick stands occur at Oregon Inlet, Hatteras 

 Village and Ocracoke where the regular tide range is less than 30 centi- 

 meters. At Beaufort, Swansboro; and Oak Island where the tide range is 

 about 1 meter, the grass is taller but the stands are sparse. 



There is also a significant difference in the number of stems per unit 

 area between the tall and short height zones. The stands are more dense 

 in the short height zones, but the yields are much less than in the tall 



107 



