Module11 Farid

8/16/2019 Module11 Farid

1/56

Sampling

Bogor, May 23, 2012

Module 11


2/56

S t e p s

i n t h e d e

v el o pm en

t of a

H S R p

r o p o s al


3/56

Objectives

At the end of the session, you should be able to:

Identify and define the population(s) to be studied

Identify and describe !ommon methods of sampling

Discuss problems of bias that should be a"oided #hensele!ting a sample

List the issues to !onsider #hen de!iding on samplesi$e

Decide on the sampling method(s) and sample si$e(s)

most appropriate for the resear!h design you arede"eloping


4/56

I. I!ROD"#!IO


5/56

$H%! is samplin&'

%he pro!ess of sele!ting a number ofstudy units from a defined study

population


6/56

Sample types

&f study population is small, total populationsample (eg All M'%B patients in the *M'%program)

&f study population is lar&e, mainly forpra!ti!al reasons, sample of the population


7/56

Samplin& (uestions

+hat is the group of people (S%'-*.*/A%&.) #e are interested in from#hi!h #e #ant to dra# a sample

o# many people do #e need in oursample

o# #ill these people be sele!ted


8/56

Study population

Should be !learly defined, for eample,a!!ording to age, se, and residen!e

4onsists of study units

5 *ersons

5 6illages

5 &nstitutions

5 *atient re!ords


9/56

Representative sample

A sample #ith all the important!hara!teristi!s of the population from#hi!h it is dra#n

)or e*ample+

&f you study health !are see7ing beha"iour in 100 %Bpatients in 8ast 9a"a you #ould ha"e to sele!t these %Bpatients from a representati"e sample of "illages!ities

&t #ould be un#ise to sele!t them from the !apital !ity asthis might gi"e you a distorted (biased) pi!ture


10/56

Definitions

Study population

5 All the sampling units or indi"iduals#hi!h !ould possibly be in!luded in the

sample Samplin& unit

5 %he item #hi!h is sampled

Samplin& interval5 %he proportion of a study population

sampled


11/56

II. S%M,LI- M!HODS


12/56

Samplin& methods

*urposi"e sampling strategies for(ualitative studies

andom sampling strategies to !olle!t

(uantitative data


13/56

,urposive samplin& strate&ies

;o!using on limited number of informants,#hom #e sele!t strategically

%heir indepth information gi"es optimal

insight into an issue about #hi!h little is7no#n

ote: not representati"e of the general

population<


14/56

,urposive samplin& strate&ies

8treme !ase sampling

Maimum "ariation sampling

omogeneous sampling

%ypi!al !ase sampling

4riti!al !ase sampling

Sno#ball or !hain sampling


15/56

*treme case samplin&

A sele!tion pro!ess that in!ludesunits #ith spe!ial or unusual!hara!teristi!s

8g &nter"ie#ing good or "ery poor !ompliers to%B treatment help in identifying !ontributing

fa!tors to poor !omplian!e


16/56

Ma*imum variation samplin&

A sele!tion pro!ess #hi!h in!ludes units sothat differen!es on spe!ified !hara!teristi!sare maimi$ed

8g ;or studying reasons for defaulting #e sele!tmales and females, poor and ri!h, indi"iduals#ith rural and urban residen!e et!

'oes not pro"ide representative data for thetotal population


17/56

Homo&eneous samplin&

A sele!tion pro!ess fo!using on parti!ularunit or subgroup !onsideredhomogeneous

8g Sele!tion of urban poor people to study theeffe!t of the introdu!tion of user fees on health!are see7ing beha"ior

sed for ;o!us =roup 'is!ussions: parti!ipantsdis!uss more freely #hen they are amongstpeople of similar so!ial status


18/56

!ypical case samplin&

A sele!tion pro!ess that in!ludes units!onsidered typi!al of the phenomenonunder study

8g Study of performan!e of a %B !lini! in a slumarea


19/56

#ritical case samplin&

Sampling of those indi"iduals #ho >!anma7e the differen!e? #ith respe!t to aninter"ention you #ant to introdu!e or toe"aluate

8g -ou #ant to introdu!e leaflets eplaining theimportan!e of nutrition for %B patients %heseleaflets !an be tested on malnourished %B

patients


20/56

Sno/ball or chain samplin&

&nter"ie# sub@e!ts are obtained fromsub@e!ts already inter"ie#ed for the study

eg &ndepth inter"ie# leads to dis!o"eries #hi!hseem re#arding to follo#up by a number ofinter"ie#s #ith an additional group of informants

.ften used in diffi!ulttorea!h populations(homeless, in@e!ting drug users, )


21/56

,robability samplin& strate&ies

&f #e #ant to &eneralise the findingsobtained from a sample to the total studypopulation

ses random sele!tion pro!edures toensure that ea!h unit of the sample is!hosen on the basis of !han!e


22/56

,robability samplin& methods

Simple random sampling

Systemati! sampling

Stratified sampling 4luster sampling

Multistage sampling


23/56

Simple random samplin&

sed in situations #here the number of samplingunits is relati"ely small

'etermine units a"ailable for sampling: ie Studypopulation of 100 indi"iduals

'e!ide on sample si$e: ie Sample 10 indi"iduals

/ottery method (random number table, !omputerprogram

01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50

51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75

76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99

100


24/56

%dvanta&es and Disadvanta&es

Ad"antages

5 8asy to understand

5 8asy to analy$e

'isad"antages

5 euires a list of the population

5 4ost may be prohibiti"e

5 May miss or undersample 7ey subsets


25/56


Ad"antages

5 /ess time !onsuming than simple randomsampling

5 8asy to perform

'isad"antages

5 is7 of bias (ie sampling days of the #ee7#ith sampling inter"al of C #ill al#ayssele!t %uesday #hi!h may be a mar7etday)


26/56

Systematic samplin&

A systemati! sample is sele!ted from 100 studentsof a s!hool %he sample si$e sele!ted is 2D

2D (E sample si$e)

100 (E study population)2D100EF, sampling inter"al

Start at random student bet#een 1 and F, eg 2

01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50

51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75

76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99

100


27/56

Stratified Samplin&

sed #hen the sampling frame !ontains!learly different !ategories (strata), eg

5 rban and rural fa!ilities

5 4ommunity and prison patients

seful if !omparisons #ill be made of theresults in subgroups of the studypopulation


28/56

Stratified samplin&

1 G 11 1G 21 2G 31 1 G 11

2 12 1 22 2 32 2 12

3 H 13 1H 23 2H 33 3 H 13F I 1F 1I 2F 2I 3F F I 1F

D 10 1D 20 2D 30 3D D 10 1D

!0 prevalence surveyural 3D "illages

rban 1D !ities

Sele!t D "illages and D !ities


29/56

Stratified samplin&

1 G 11 1G 21 2G 31 1 G 11

2 12 1 22 2 32 2 12

3 H 13 1H 23 2H 33 3 H 13

F I 1F 1I 2F 2I 3F F I 1F

D 10 1D 20 2D 30 3D D 10 1D

!0 prevalence survey

ural 3D "illages

rban 1D !ities

Sele!t D "illages and D !ities by simple randomsampling


30/56

Stratified samplin&

1 G 11 1G 21 2G 31 1 G 11

2 12 1 22 2 32 2 12

3 H 13 1H 23 2H 33 3 H 13

F I 1F 1I 2F 2I 3F F I 1F

D 10 1D 20 2D 30 3D D 10 1D

!0 prevalence surveyural 3D "illages

rban 1D !ities

Sele!t D "illages and D !ities by systemati! sampling


31/56


Ad"antages

5 4an dra# !on!lusions about relati"elysmall group

'isad"antages

5 More time !onsuming

5 neual sampling fra!tions: must be

!orre!ted for if generali$ing findings to#hole population


32/56

#luster Samplin&

sed #hen for logisti! reasons it is easierto sele!t sample units in groups5 =eographi!: distri!ts"illages5 .rganisational units: s!hools!lini!s

Method5 &dentify !lusters of units of interest5 Sele!t !lusters of sample units randomly


33/56

#luster samplin&

1 2 3 F D G H I 10

11 12 13 1F 1D 1G 1 1H 1I 2021 22 23 2F 2D 2G 2 2H 2I 30

31 32 33 3F 3D 3G 3 3H 3I F0

F1 F2 F3 FF FD FG F FH FI D0

%uber!ulin sur"ey #ith 1,000 !hildren

&n total D0 s!hools in the region

Sele!t 10 s!hools

Sele!t 100 !hildren per s!hool


34/56

#luster samplin&

1 2 3 F D G H I 10

11 12 13 1F 1D 1G 1 1H 1I 2021 22 23 2F 2D 2G 2 2H 2I 30

31 32 33 3F 3D 3G 3 3H 3I F0

F1 F2 F3 FF FD FG F FH FI D0

%uber!ulin sur"ey #ith 1,000 !hildren

&n total D0 s!hools in the region

Sele!t 10 s!hools by simple random sampling

Sele!t 100 !hildren per s!hool


35/56

Multista&e samplin&

Sample si2e 1345333 individualsStratifi!ation of distri!ts (urban, rural, remote)

Sorting all GIF distri!ts by population si$e

Stage 1, systemati! sampling of 0 distri!ts: 20urban, 30 rural, and 20 remote distri!ts

Stage 2, random sele!tion of 1 !ommune perdistri!t

Stage 3, random sele!tion of 1 sub!ommuneper !ommune


36/56

0ias due to improper samplin&

Studying "olunteers only Sampling of registered patients only Missing !ases of short duration

Seasonal bias %arma! bias (a!!essibility)

onresponse


37/56

III. S%M,L SI6


38/56

Sample si2e

%he bigger the better

Sample si$e depends on:

5 8pe!ted "ariation in the data

5 4onfiden!e inter"al pre!ision needed

5 Si$e of target population

%he e"entual sample si$e is usually a

!ompromise bet#een #hat is '8S&AB/8and #hat is ;8AS&B/8


39/56

Sample si2e calculations

Measuring one single "ariable in one group #ith a!ertain pre!ision (ie a mean, a rate or a proportion)

%rying to demonstrate a signifi!ant differen!e

bet#een groups

se of formula or table for !al!ulating sample si$e


40/56

$hen to #alculate the Sample

Si2e &t is important to estimate the sample si$e

early in the design phase

+aiting until the last minute (or after thestudy


41/56

Hypothesis !estin& ull hypothesis (o): %he proportion of

tuber!ulosis among smo7ers is notdifferent from nonsmo7ers

Alternati"e hypothesis (A): %he

proportion of tuber!ulosis among smo7ers

is greater than among nonsmo7ers (onesided)J . is different than among nonsmo7ers (t#osided)


42/56

!ype I rror 7 probability of !ype I error (falsepositi"e error

rate)

8"en if the null hypothesis is %8, #e #ill, inrepeated samples, re@e!t o a proportion 8 of the

time

%raditionally, 8 9 3.34 5 the probability ofma:in& a !ype I error

5 %hus, the li7elihood of erroneously re@e!ting the nullhypothesis is DK

1 L E IDK !onfiden!e le"el


43/56

!ype II rror ; ,o/er


44/56

!ype II rror ; ,o/er

%raditionally, < 9 3.?35 the probabilityof ma:in& a !ype II error

%hus, there is a 20K !han!e of ma7ing a

%ype && error and failing to re@e!t the nullhypothesis #hen the alternati"e is true

%herefore, the po#er E (1020) E 0H0 orH0K (!han!e of dete!ting a differen!e of

the magnitude spe!ified if one truly eists)


45/56

$hat else do /e need'

8pe!ted differen!e bet#een 2 groups

.ther, similar studies

*ilot study

An informed best guessN or estimate ofan important differen!eN

8pe!ted "ariability in the measurements


46/56

)ormula+ #omparin& ?proportions


47/56

Sample si2e in (uantitative studies

#omparin& t/o &roups for a si&nificant difference

%o !ompare the differen!e in B4= immuni$ation !o"eragebet#een t#o regions

1 8stimate !o"erage in both regions (say G0K (p0) and H0K

(p1))2 4hoose the li:elihood =study po/er> that the

"a!!ination !o"erage in the t#o regions is indeed different(say I0K)

3 nE (G0 x F0OH0 x 20) x (12HO1IG)2 E 10D

(H0G0) 2

#onf.

level

2 =1 8> po/er 2 =1

I0K 1GFD H0K 0HF

IDK 1.@A I0K 12H

(a)

Sample size for comparison of proportions in two groups


48/56

Percentage 1

10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

Percentage 2

5% 577 97 43 25 15 10

10% 262 79 38 22 14

15% 913 1207 157 62 32 19 1220% 262 388 105 48 26 16

25% 129 1459 1669 199 73 37 21 12

30% 79 388 472 121 52 28 16

35% 53 181 1837 1963 223 79 37 20 11

40% 38 105 472 514 126 52 26 14

45% 29 68 213 2047 2089 227 77 35 17

50% 22 48 121 514 514 121 48 22 10

55% 17 35 77 227 2089 2047 213 68 29 13

60% 14 26 52 126 514 472 105 38 16

65% 11 20 37 79 223 1963 1837 181 53 19

70% 16 28 52 121 472 388 79 24

75% 12 21 37 73 199 1669 1459 129 31

80% 16 26 48 105 388 262 4285% 12 19 32 62 157 1207 913 59

90% 14 22 38 79 262 94

95% 10 15 25 43 97 577 199

100% 10 16 24 42 94

u= 1.28 Power=90%

v= 1.96 Significance: P


49/56

)ormula+ sin&le proportion

n E P2 p

d2

P E 1IG, the normal "alue !orresponding to theIDK !onfiden!e inter"al

* E 0H0, pre"alen!e from the abo"e study

E 1p E 020

d E margin of error E desired pre!ision E 002D

n E 1IG2 (0H Q 02) E IHF 0,02D2


50/56

Other Increases ecessary in

Sample Si2e Determination

emember to a!!ount for:

nonparti!ipation (usually estimatedat 101DK)

loss to follo#up (in a !ohort study)#hi!h !an be substantial

other issues su!h as inability to find!harts, et!


51/56


52/56

sampsi 0G 0D, alpha(00D) po#er(0I0)


53/56

http:sampsi$esour!eforgenet

http://sampsize.sourceforge.net/http://sampsize.sourceforge.net/http://sampsize.sourceforge.net/


54/56

-RO", $ORB


55/56

=.* +.R, *A% & (2 hours)

'e"elop in your #or7ing group:

a definition of your (different) study population(s)J

a definition of your (different) study units (people, !lini!s,re!ords, et!)J

appropriate sampling pro!edures for your study, ta7inginto a!!ount #hether you use ualitati"e andoruantitati"e resear!h methods State ho# you #ill try toa"oid possible bias

*repare a summary on a flip!hart for use in the eer!ise

>4ommenting on ea!h others sampling pro!edures? and inthe plenary dis!ussion (after group #or7 on sample si$e)


56/56

=.* +.R, *A% && (1 hour)

'etermine the sample si$e reuirements for the studypopulation(s) defined in the pre"ious group #or7 session

'etermine the feasible sample si$e after ta7ing intoa!!ount a"ailable time, manpo#er, transport and money

&f there is a large dis!repan!y bet#een the desirable andthe feasible sample si$e you should loo7 for a!ompromise and, if ne!essary, ad@ust the ob@e!ti"es ofyour study

*ut a summary of your groups #or7 on flip!hart for use

in the eer!ise belo# and in the plenary dis!ussion that#ill follo#

Module11 Farid

Documents