Table 2— Results of analyses of variance for the periodicity of shoot elongation presented as intra- 

 class correlations, the ratio of a variance component to the sum of all components 



Variable 



Source of 



variance 



Elongation 



Initiation 



Cessation 



Duration 



Rate 



Replications 



0.04** 



0.03** 



0.01 *- 



0.01** 



0.07** 



Populations 



.20** 



.02'* 



.34** 



.17** 



.15** 



Interaction 



.05** 



.09** 



.16'* 



.03** 



.06** 



Within plots 



.71 



.86 



.49 



.79 



.71 



Error mean square' 



766.931 



4.811 



18.109 



22.553 



1.428 



"Significance of the F-value at the 1 percent level of probability. 



'Absolute value of the interaction mean square that was used to test for differences between populations. 



Table 3 — Results of stepwise regressions for relating genetic variation to the eleva- 

 tion and geographic origin of the seed 



Number of 



independent Residual 

 Variable variables mean square 



Growth and development 



3-year mesic height 



5 



0.52" 



10.09 



3-year xeric height 



6 



.31" 



4.55 



Adjusted mesic height 



6 



.17" 



2.14 



Adjusted xeric height 



6 



.13" 



1.91 



Leaf length 



5 



.10" 



.20 



Xeric height reduction 



8 



.40** 



6.20 



eriodicity of shoot elongation 









Elongation 



5 



.49** 



86.98 



Initiation 



5 



.09 



.24 



Cessation 



6 



.59" 



1.74 



Duration 



5 



.58" 



1.89 



Rate 



6 



.43" 



.14 



"Statistically significant at the 1 percent level. 



available for the logistic regression that described shoot 

 elongation of individual trees nearly perfectly. Values of 

 ranged from 0.90 to essentially 1.0, averaging 0.99. 

 Analyses of variance not only detected significant dif- 

 ferences between populations for all variables (table 2) but 

 also attributed substantial variance to the effects of popu- 

 lations for all variables except the initiation of shoot elon- 

 gation. These effects were reflected in mean differences 

 between populations as large as 9 days in cessation, 

 10 days in duration, 2.8 mm per day in rate, 73 mm total 

 elongation, but only 2.3 days in initiation. Although these 

 differences seem small, culturing plants under optimal con- 

 ditions allows shoot development to proceed rapidly, and 

 developmental events are condensed into a short time 

 interval. Under natural conditions development occurs 

 over a long period and differences in the periodicity of 

 shoot elongation become relatively large. 



Patterns of Genetic Variation 



By accounting for as much as 60 percent of the variance 

 between populations, regression models were statistically 

 significant for all variables except the initiation of shoot 



elongation (table 3) and drought mortality, two variables 

 for which population differentiation was not pronounced 

 (tables 1 and 2). Thus, genetic differences between popula- 

 tions follow systematic patterns that can be related to the 

 elevation and geographic origin of the seed. These pat- 

 terns can be presented as (1) elevational clines for several 

 geographic localities (fig. 2), (2) geographic clines at a con- 

 stant elevation (fig. 3), or (3) geographic clines at the base 

 elevation, the lowest elevation at which the species occurs 

 in a given locality (fig. 4). 



These results, however, were based on five to eight in- 

 dependent variables and, therefore, are susceptible to 

 overfitting, a condition that occurs when models are fit to 

 single samples rather than to general trends described by 

 all samples (Draper and Smith 1981). Consequently, in this 

 paper, population differentiation is interpreted according 

 to the least significant difference (Steel and Torrie 1960) 

 at the 80 percent level of probability— Zs(i(0. 2). A relatively 

 low level of probabDity is used to guard against accepting 

 no differences between populations when differences ac- 

 tually exist. Thus, in figure 2, populations separated by an 

 elevational interval sufficient to subtend a mean difference 

 equal to lsd{0.2) are judged to be statistically different at 



4 



