200 8
Determining Sample Size
Two approaches to tackle the issue
By Raja Rub Nawaz
Statistical metho s !or etermining sample size
"robability sample sizes shoul be etermine by calculation# There are lots o! te$tbooks which actually eal with this matter in etail# % un erstan that sometimes either those books contain lengthy iscussion on this topic or many times they eal with it &ery technically# 'y approach to tackle this issue is &ery simple# (et)s begin our iscussion a&oi ing any serious an &ery technical calculations# There are two i!!erent types o! calculation that can be one# *&erage +or mean, metho "roportion +or percentage, metho
The i!!erent types o! calculation are escribe in the !ollowing sections#
Calculations involving means
The !ormula which is use !or calculating the size o! a ran om sample which is to be use to in&estigate a mean +a&erage, is-
.here
The three pieces o! in!ormation nee e to calculate /n01 are-
1. An acceptable level of random error (E)
This !igure speci!ies the e$tent to which the researcher is willing to tolerate the !in ings !rom the sample being i!!erent !rom the entire population# The higher the le&el o! /201 the less chances are there that !in ings !rom the 3uestionnaire +sur&ey, represent the &iews o! the population#
%! we a or minus this number to the a&erage or mean1 it gi&es us the range o! our con!i ence inter&al !or the mean# 4or e$ample1 suppose a cinema was to con uct a sur&ey to etermine how o!ten each year1 on a&erage1 people atten the cinema# The researcher may choose a le&el o! sampling error +2, o! 0#2# %! the a&erage number o! &isits by the respon ents in the sample is !oun to be 5#6 per year1 the researcher can now easily !in that the a&erage number o! &isits is anywhere between 5#5 an 5#7 time per month1 i#e# 5#6 8 0#2# This range is known as the con!i ence inter&al !or the mean#
2. An acceptable confidence level (Z)
This !igure tells us the chances that the !in ings !rom the sample will !all within the con!i ence inter&al# %n the !ormula !or calculating the size o! a sample +gi&en abo&e, con!i ence le&els are e$presse as 9:) &alues# The most commonly use 9:) &alues shown in table ; below#
Table 1: Most commonl used Z values
Confidence !evel (") #$ #% ##
Associated Z value ;#<=5 ;#7<0 2#565
&. An estimate of t'e standard deviation of t'e population (()
%n cru e terms this !igure can be thought o! as the e$tent to which members o! population are similar or i!!erent to each other +this is why it is calle / ispersion0,# The lower the stan ar e&iation1 the more similar its members are to each other an the higher the stan ar e&iation1 the less similar + i!!erent, are the members to each other# *n estimate o! the stan ar e&iation o! a population is usually ma e on the basis o! pre&ious sur&eys1 pilot sur&eys1 secon ary ata1 or simply the ju gment o! the researcher#
An E)ample of calculations involving MEA*+: * sur&ey may be esigne to in&estigate how o!ten1 on a&erage1 people atten their local cinema# (et)s assume that the three pieces o! in!ormation are as !ollows*n acceptable le&el o! ran om error +2, > 0#2 *n acceptable con!i ence le&el +:, > 75?1 so we will use the : &alue as ;#7<0 !rom table ; abo&e# *n estimate o! the stan ar e&iation o! the population +@, > ;#56
.hen these three pieces o! in!ormation ha&e been generate 1 the calculation o! an appropriate sample size is &ery easy#
Base on abo&e calculations1 a simple ran om sample o! 2A< people is re3uire to pro&i e the researcher with the esire le&els o! accuracy with 75? con!i ence#
Calculations involving proportions
The !ormula which is use !or the calculating the size o! a ran om sample use to in&estigate a proportion is-
.here
The three pieces o! in!ormation nee e to calculate /n01 are-
1. An acceptable confidence level (Z)
This !igure can be selecte in precisely the same way as escribe !or calculations in&ol&ing means as escribe earlier#
2. An estimate of t'e population proportion (,)
The estimate the population proportion +",1 the researcher has to guess this#
&. An acceptable level of random error (E)
This !igure speci!ies the e$tent to which the researcher is willing to tolerate the !in ings !rom the sample being i!!erent !rom the entire population . -n t'is scenario of
calculations involving proportions. E /ill be e)pressed as a percentage rat'er t'an an absolute value. 4or e$ample1 suppose a cinema was to con uct a sur&ey to in&estigate the proportion o! people who atten more than once a month# The researcher may choose a le&el o! sampling error +2, o! 0#0A an the proportion o! people &isiting the cinema more than once a month is !oun to be ;<#5?1 then the researcher can now easily !in that the people who atten the cinema more than once a month is anywhere between ;A#5? an ;7#5?1 i#e# ;<#5? 8 A?# This range is known as the con!i ence inter&al !or the proportion# An E)ample of calculations involving ,01,10T-1*+: * sur&ey may be esigne to in&estigate the proportion o! people who &isit their local cinema more than once a month# (et)s assume that the three pieces o! in!ormation are as !ollows*n acceptable le&el o! ran om error +2, > 0#0A *n acceptable con!i ence le&el +:, > 75?1 so we will use the : &alue as ;#7< !rom table ; abo&e# *n estimate o! the population proportion +", atten ing the cinema more than once a month > 0#;5 +or ;5?, .hen these three pieces o! in!ormation ha&e been generate 1 the calculation o! an appropriate sample size is calculate as !ollows-
Base on abo&e calculations1 a simple ran om sample o! 5=; people is re3uire to pro&i e the researcher with the esire le&els o! accuracy with 75? con!i ence#