Determining Sample Size WWW Resources

Determining Sample Size: Balancing Power, Precision, and Practicality

Internet Resources for Determining Sample Size Organized by Chapter

Chapter 1

Chapter 2

Chapter 3

Chapter 4

Chapter 5

Chapter 6

Appendix

Additional Resources

Book Image

published by Oxford University Press

Sample size determination is an important and often difficult step in planning an empirical study. From a statistical perspective, sample size depends on the following factors: type of analysis to be performed, desired precision of estimates, kind and number of comparisons to be made, number of variables to be examined, and heterogeneity of the population to be sampled. Other important considerations include feasibility, such as ethical limitations on access to a population of interest and the availability of time and money. The primary assumption of this book is that, within the context of ethical and practical limitations, efforts to obtain samples of appropriate size and quality remain an important and viable component of social science research.

This text describes the following available approaches for estimating sample size in social work research and discusses their strengths and weaknesses: (1) power analysis, (2) confidence intervals, (3) computer-intensive strategies, and (4) ethical and cost considerations. In addition, strategies for mitigating pressures to increase sample size, such as emphasis on model parsimony (e.g., fewer dependent and independent variables), simpler study designs, an emphasis on replication, and careful planning of analyses are discussed. This text covers sample-size determination for advanced and emerging statistical strategies, such as structural equation modeling, multilevel analysis, repeated measures MANOVA, and repeated measures ANOVA which are not discussed in other texts including Cohen (1988). This book can serve either as a supplemental text for an introductory course or as a core text for a more advanced course.

This page contains two calculators. This page contains calculators for the normal distribution. These calculators can be used instead of a table of the normal distribution. (p. 44)

Calculator for the confidence interval for the difference between two independent proportions (p. 45)

WHATIS.EXE calculates confidence intervals for a variety of statistics, probability and inverse probability values, and time-spans. (p. 45)

Noncentral confidence interval and power calculations (SPSS scripts) (p. 46)

Calculate for confidence interval around an odds ratio (p. 46)

Calculate confidence intervals around chi square (p. 46)

Noncentral confidence interval and power calculations (SPSS scripts) (p. 48)

PSY calculate confidence intervals around contrasts (p. 49)

Statistica power analysis module (p. 49)

Calculate confidence intervals around rho (p. 49)

Calculate the 99%, 95%,and 90% confidence intervals for a regression coefficient (p. 51)

Calculate the 99%, 95%,and 90% confidence intervals for R-squared (p. 51)

Calculate an adjusted R-squared (p. 51)

R2 calculates confidence intervals for R-squared (p. 51)

R2 manual [pdf] (p. 51)

Statsoft (Statistica) (p. 54)

Chapter 4

DataSim generates data that conform to user specifications (p. 63)

Random Number Generator (p. 63)

Simstat is a program for data analysis and simulation (p. 64)

David Howell's Resampling Statistics page (randomization and the bootstrap) (p. 66)

The following syntax generates 2,000 cases with three normally distributed variables:

set seed = 200409281.
input program.
loop #i = 1 to 2000.
compute x_normal = rv.normal(50,15).
compute y_normal = rv.normal(25,7).
compute z_normal = rv.normal(5,2).
end case.
end loop.
end file.
end input program.
formats x_normal (f8.4).
execute.
descriptives variables = x_normal y_normal z_normal /statistics = all.

The SPSS syntax file for example on pp. 64-65 is as follows:

DEFINE repsamp ().
!DO !doover = 1 !TO 40.
USE ALL.
do if $casenum = 1.
compute #s_$_1=40.
compute #s_$_2=1000.
end if.
do if #s_$_2 > 0.
compute filter_$ = uniform(1)* #s_$_2 < #s_$_1.
compute #s_$_1 = #s_$_1—filter_$.
compute #s_$_2 = #s_$_2—1.
else.
compute filter_$ = 0.
end if.
VARIABLE LABEL filter_$ ’10 from the first 1000 cases (SAMPLE).’
FORMAT filter_$ (f1.0).
FILTER BY filter_$.
T-TEST
GROUPS = group(1 0)
/MISSING = ANALYSIS
/VARIABLES = posttest
/CRITERIA = CI(.95) .
!DOEND.
!ENDDEFINE.
repsamp.
execute.