چه تعداد تکرار بوتسترپ برای بررسی‌های جدول زندگی ضروری است؟

An important problem in life table analysis is estimating uncertainty of the parameters. Researchers use recombination (or resampling) methods, while bootsrap is preferred recently to jackknife technique. One basic question in these analyses is that how many bootstrap replication is necessary to achieve an adequately precise estimate without unnecessary measurements. In recent researches 100'000 replicates are used, however Meyer et al proposed 500-1000 replicates. In this study, we tackled this issue by addressing two basic objects; one normality of parameters; i. e. if higher number of replicates lead to achieve a normal distribution, and the second, how precise is standard error (SE) estimates. If additional replicates do not play a role in achieving more precise and normal parameter estimates, then they will be avoided. To achieve this task, we processed data of 19 cohorts including six Trissolcus vassilievi Mayr, of BenaMolaei’s research, two cohorts of Ooencyrtus telenomicida and O. fecundus and 11 simulated life table data, and 10 or 100 thousands bootstrap replicates were created by computer and trend of convergence of SE values by including any additional replicate, as well as normality of data were investigated. Taking into account that laboratory data had only convex survivorship curves, in simulated populations some further variables such as concave vs. convex survivorship curve, skewed vs. normal distribution of reproduction, higher vs. lower growth rates and developmental time, also were included. Among those variables only immature mortality (concave vs convex survivorship curve) had a considerable effect on the behavior of parameters. As long as survivorship curve is convex, d remains skewed and did not reach a normal situation, while 500-1000 replicates was enough to achieve convergence in SE-values. Additional replicates did not lead to achieve normal data. In contrast when the survivorship curve was concave, d tend to achieve to normality, whereas the other parameters depart more frequently. Convergence in SE estimates achieved by higher number of replicates particularly when number of the handled insects were low. In this circumstances 1600-2500 replicates caused SE-values to converge. As a whole, bootstrap replicates over 1000 when immature mortality is low, and 2500 when it is high adequate and additional replicates are unnecessary and do not assist in obtaining more precise and normal set of data.