Curs 3 CB Metodologia Cercetarii

8/19/2019 Curs 3 CB Metodologia Cercetarii

1/98

Corin Badiu, 2007

Metodologia cercetMetodologia cercetăăriirii şştiintiinţţificeifice

Colectarea datelor

Strategii de masurare


2/98


3/98


4/98

Tipuri de studii clinice

• Studii fara variabile♦ Studii de caz serii de cazuri editoriale

opinii ) comentarii rapoarte revie*• Studii cu o singura variabila

♦ Studii descriptive

• Studii cu +, variabile♦ "#perimente♦ Studii observationale♦ Meta-analize si revie*-uri sistematice


5/98


6/98


7/98

"vidence /ased Medicine

• Metode de tratament sustinute dedovezi clinice si de cercetare0

• 1ecesita integrarea celor mai bunedovezi din cercetare pentru diagnosticsi tratament cu e#perienta clinica0

• Ia in considerare ce este optim pentrufiecare pacient precum si preferinteleacestuia0


8/98

'ealitatea

• Cercetarea asupra eficientei de tratamentface subiectul unui numar mic de articole0

• Evidence based medicine este considerat unconcept ce foloseste baze de date inclusivstudii sistematice de caz pentru a g.idainterventii terapeutice0

• 2ovezile trebuie evaluate intr-un conte#tterapeutic efectiv 3Ce tip de interventiecapata sens pentru mine ca practician45


9/98

Clinical 6uestions

• 7.at is t.e best c.oice of t.erap8

for m8 patient4• Is t.is program t.eoreticall8 sound4

• 2oes t.is t.erap8 program *or94

• :o* long *it. t.e t.erap8 ta9e4

• 7.ere do I go from .ere4


10/98

$racticing Clinicians 1eeded;

Clinicians

♦ are on t.e front line

♦ .ave necessar8 clinical e#pertise

♦ 9no* t.eir patients *ell

♦

are naturall8 scientific t.in9ers♦ are *ell-versed in data collection

♦ 9no* .o* to loo9 for outcomes


11/98

/aseline Measurements

• A baseline is a measure of responserates in t.e absence of treatment

• /aselines

♦ "stablis. a need for treatment

♦

2ocument improvement♦ Allo* us to modif8 if *e don


12/98

/aseline 2ata

• Create a set of e#emplars of eac. of 8our

targets and prepare a recording s.eet0• =tilize criterion referenced measures0


13/98

2ata Collection Strategies

• Al*a8s .ave more t.an one measurement

• C.ec9 t.e reliabilit8 of t.e baseline data• Select researc.)clinical design


14/98

'esearc.)Clinical 2esigns

• A/A designs♦ Test treat and test

• A/A/ designs♦ Test treat test and treat

• Time-Series designs♦ "stablis. stable baseline♦ /egin treatment♦ Measure treatment results

• Multiple-/aseline designs♦ :ave a number of different baselines♦ "ac. baseline must be independent of t.e ot.ers♦ Onl8 treat one variable


15/98

2ata Collection Instruments

Requirements

• 'eliable

• (alid

• 'esponsive• =niversal

• =nbiased


16/98

2ata Collection Instrument

Is it reliable?

7ill t.e instrument measure consistentl8 across!

• 2ifferent testing situations4Test-retest reliabilit8

• 2ifferent >udges4

Inter-rater reliabilit8


17/98


Is it valid?

• Is t.e instrumentbeing used tomeasure t.e 9ind of

data for *.ic. it*as intended4


18/98


Is it responsive?

• T.e instrument s.ould be e?uall8 sensitive *.et.er a

c.aracteristic is present or absent0Must measure both

• %alse-negatives!

@ou t.oug.t it *as intact but it *as torn0

• %alse-positives! @ou t.oug.t it *as torn but it *as intact0


19/98


Is it universal? T.e investigator s.ould

emplo8 a *idel8 used datacollection instrument *.ic..elps minimize reporting biasbecause t.e data can t.en be

compared *it. ot.er publis.edliterature0


20/98


• T.ere s.ould be no difference bet*een t.e true valueand t.e value t.at an investigator actuall8 obtains

ot.er t.an a difference caused b8 sampling variabilit80

!s it unbiased"


21/98

Samplin# Methods

• Random Sampling (Simple)

• Systematic Sampling

• Stratified Sampling

4. Cluster Sampling

5. Convenience Sampling

6. More complex sampling


22/98

$ualitative and $uantitative Variables

Examples of ualitative varia!les are occupation" sex" marital

status" and etc

#aria!les t$at yield o!servations t$at can !e measured are

considered to !e uantitative varia!les. Examples of

uantitative varia!les are %eig$t" $eig$t" and age

&uantitative varia!les can furt$er !e classified as discrete orcontinuous


23/98

Variables t%pes

&' Cate#orical variables (e'#' Sex Marital Status

income cate#or%)

' Continuous variables (e'#' A#e income*ei#ht hei#ht time to achieve an outcome)

+' Discrete variables (e'#',umber o Children in

a amil%).' /inar% or Dichotomous variables (e'#'

response to all 0es or ,o t%pe o 1uestions)


24/98

Scale o Data

'. ominal *$ese data do not represent an amount or uantity (e.g."

Marital Status" Sex)

+. ,rdinal *$ese data represent an ordered series of relations$ip (e.g."level of education)

-. nterval *$ese data is measured on an interval scale $aving eual

units !ut an ar!itrary /ero point. (e.g. *emperature in 0a$ren$eit)

4. nterval Ratio #aria!le suc$ as %eig$t for %$ic$ %e can compare

meaningfully one %eig$t versus anot$er (say" '11 2g is t%ice 51 2g)


25/98

Variables in the protocol

• 203ES O4 VA5!A/6E♦ independent

♦ dependent

♦

intermediate♦ conoundin#


26/98

!ndependent Variable

• 2he characteristic bein# observed and7ormeasured that is h%pothesized to inluence an

event or outcome (dependent variable)'

• ,O2E

♦ The independent variable is not influenced

by the event or outcome, but may cause it

or contribute to its variation.


27/98

Dependent Variable

• A variable *hose value is dependent onthe eect o other variables (ie'

8independent variables9) in therelationship bein# studied' S%non%ms

outcome or response variable'

• ,O2E

♦ an event or outcome whose variation we

seek to explain or account for by the

influence of independent variables.


28/98

!ntermediate Variable

• A variable that occurs in a causal path*a% roman independent to a dependent variable'

S%non%ms intervening, mediating • ,O2ES

♦ it produces variation in the dependent

variable, and is caused to vary by the

independent variable.

♦ such a variable is “associated” with both the

dependent and independent variables.


29/98

Conoundin# Variable

• A actor (that is itsel a determinant o theoutcome) that distorts the apparent eect o a

stud% variable on the outcome'

• ,O2E

♦ such a factor may be unequally distributed

among the exposed and the unexposed, and

thereby influence the apparent magnitude and

even the direction of the effect.


30/98

Or#anizin# Data

'. 0reuency *a!le

+. 0reuency 3istogram

-. Relative 0reuency 3istogram4. 0reuency polygon

5. Relative 0reuency polygon

6. ar c$art

. ie c$art7. stem8and8leaf display

9. ox lot


31/98

4re1uenc% 2able

Suppose %e are interested in studying t$e num!er of

c$ildren in t$e families living in a community. *$e

follo%ing data $as !een collected !ased on a randomsample of n : -1 families from t$e community.

+" +" 5" -" 1" '" -" +" -" 4" '" -" 4" 5" " -" +" 4" '" 1" 5" 7" 6"

5" 4 " +" 4" 4" " 6,rgani/e t$is data in a 0reuency *a!le;


32/98

B1o0 of C.ildren Count

D%re?0E

'elative %re?0

F , ,)GFF0FH

J G G)GFF0JFF, K K)GFF0JH

G K K)GFF0JH

L H H)GFF0,FF

K L L)GFF0JGGH , ,)GFF0FH

, ,)GFF0FH

J J)GFF0FGG

4re1uenc% 2able


33/98

4re1uenc% plot


34/98

o% suppose %e need to construct a similar freuency ta!le for t$e

age of patients %it$ 3eart related pro!lems in a clinic.

*$e follo%ing data $as !een collected !ased on a random sample of

n : -1 patients %$o %ent to t$e emergency room of t$e clinic for

3eart related pro!lems.

*$e measurements are 4+" -7" 5'" 5-" 41" 67" 6+" -6" -+" 45" 5'" 6"

5-" 59" 4" 6-" 5+" 64" 6'" 4-" 56" 57" 66" 54" 56" 5+" 41" 55" +" 69.

4re1uenc% 2able


35/98

Age Nroups %re?uenc8 'elative%re?uenc8

G, -GH0 , ,)GFF0FHG- LJ0 G G)GFF0JFF

L,-LH0 L L)GFF0JGL

L-KJ0 G G)GFF0JFF

K,-KH0 )GFF0,HK-HJ0 G G)GFF0JFF

H,-HH0 L L)GFF0JGL

H-, G G)GFF0JFF

Total nGF J0FF

4re1uenc% 2able


36/98

Measures o Central 2endenc%


37/98

Empirical 5ule

0or a ormal distri!ution approximately"

a) 67> of t$e measurements fall %it$in one standard

deviation around t$e mean

!) 95> of t$e measurements fall %it$in t%o standard

deviations around t$e mean

c) 99.> of t$e measurements fall %it$in t$ree

standard deviations around t$e mean


38/98

$rere?uisite S9ills

• %undamental concepts of measurement• Scales of measurement

• 2istribution central tendenc8 variabilit8probabilit8• 2isease prevalence and incidence• 2isease outcomes Deg fatalit8 ratesE

• Associations Dcorrelation or covarianceE• :ealt. impact Deg ris9 differences and ratiosE• Sensitivit8 specificit8 predictive values


39/98

Scales of Measure

• ,ominal ?ualitative classification of e?ualvalue! gender race color cit8

• Ordinal - ?ualitative classification *.ic. canbe ran9 ordered! socioeconomic status offamilies

• !nterval - 1umerical or ?uantitative data! canbe ran9 ordered and sizes compared !temperature

• 5atio - interval data *it. absolute zero value!time or space


40/98

2istribution Central Tendenc8PMean


41/98

P(ariabilit8 $robabilit8P

• Mean

• Median

• Mode

• Standard deviation

•Statistical Significance p Q 0FJ


42/98

Confidence Interval


43/98

Statistical Significance

T8pe I and T8pe II errors

:o True :o %alse

'e>ect :o T8pe I error Correctdecision

2o 1ot 'e>ect:o

Correctdecision

T8pe II error

1ull :8pot.esis :o


44/98

Statistics Online Te#tboo9

• T.e Statistics :omepage

• .ttp!))***0statsoftinc0com)te#tboo9)sta

t.ome0.tml

http://www.statsoftinc.com/textbook/stathome.htmlhttp://www.statsoftinc.com/textbook/stathome.htmlhttp://www.statsoftinc.com/textbook/stathome.htmlhttp://www.statsoftinc.com/textbook/stathome.html


45/98

2isease $revalence and

Incidence• $revalence♦ probabilit8 of disease in entire population at

an8 point in time♦ ,R of t.e population .as diabetes

• Incidence

♦ probabilit8 t.at patient *it.out diseasedevelops disease during interval

♦ F0,R or , per JFFF ne* cases per 8ear


46/98

Sensitivit8 Specificit8

• sensitivit% a ) DacE

• speciicit% d ) DbdE

3atients*ithdisease

3atients*ithoutdisease

2est ispositive a b

2est isne#ative

c d


47/98

$redictive (alue

• 3ositive predictive value a ) D abE

• ,e#ative predictivevalue d ) DcdE

• 3ost-test probabilit% odisease #iven positivetest a ) DabE

• 3ost-test probabilit% odisease #iven ne#ative

test c ) DcdE

3atients*ithdisease

3atients*ithoutdisease

2est ispositive a b2est isne#ative

c d


48/98

Nood 'esource Sen Spc $(

• An Introduction to Information Master8

• .ttp!))***0poems0msu0edu)InfoMaster8)defau

lt0.tm ♦ 2iagnosis

• Sensitivit8 and specificit8

• $redictive values• i9eli.ood ratios

• Info'etriever♦ Calculators! "pidemiolog8 2iagnostic test

http://www.poems.msu.edu/InfoMastery/default.htmhttp://www.poems.msu.edu/InfoMastery/default.htmhttp://www.poems.msu.edu/InfoMastery/default.htmhttp://www.poems.msu.edu/InfoMastery/default.htm


49/98

/ias in Clinical Trials

Areas in which bias can occur

S8stematic error in 0 0 0

• Allocation• 'esponse

• Assessment


50/98


Allocation or Susceptibility Bias

• Can occur *.en patient assignments to a trial

group are influenced b8 an investigator


51/98


Allocation or Susceptibility Bias

• Treatment groups must .ave similarprognoses *.ic. is ac.ieved b8!♦ 'andomization of patients

♦ $rospective evaluation of patients

♦ 7ell-defined inclusion and e#clusion criteria


52/98

'andomization in Clinical Trials

Occurs *.en patientsare assigned totreatments b8 means

of a mec.anism t.atprevents bot. t.epatients and t.einvestigator from9no*ing *.ic.

treatment is beingassigned0


53/98

/enefits of 'andomization• $revents t.e s8stematic introduction of bias0

• Minimizes t.e possibilit8 of allocation bias0

• /alances prognostic factors for treatment groups0

• Improves t.e validit8 of statistical tests used to

compare treatments0


54/98


Response & Assessment/Recording Bias

• Can occur *.en a patient reports a treatment

response or *.en an investigator assesses t.atresponseUeit.er person can be influenced b89no*ing t.e treatment0

• A patient or an investigator ma8 .ave apreconceived idea of *.ic. treatment is better0T.e patient ma8 also *ant to please t.einvestigator0


55/98


Blinding

To minimize 'esponse V Assessment)'ecording /ias

• Single /lind Dpatient blindedE! protects againstresponse bias0

•2ouble /lind Dpatient and investigator blindedE!protects against assessment)recording bias as *ellas response bias0


56/98


rans!er bias

♦ Occurs *.en patients are lost to follo*-up0

♦ Must be minimized0

"er!ormance bias

♦ Can occur *it. a single surgeon or *it.multiple surgeons0


57/98

Confounding "#ample

• 'elations.ip bet*een coffee andpancreatic cancer /=T

• Smo9ing is a 9no*n ris9 factor forpancreatic cancer

• Smo9ing is associated *it. coffee

drin9ing but it is not a result of coffeedrin9ing0


58/98

7.at is confounding4

• If an association is observed bet*eencoffee drin9ing and pancreatic cancer

♦ Coffee actuall8 causes pancreatic canceror

♦ T.e coffee drin9ing and pancreatic cancerassociation is t.e result of confounding b8cigarette smo9ing0


59/98

:o* to .andle confounding

• If 8ou 9no* somet.ing is a possibleconfounder in t.e data anal8sis use

♦ Stratification or ♦ Ad>ustment

• %ear t.e un9no*n;


60/98

Stud8 2esign Ta#onom8

• Treatment vs0 Observational

• $rospective vs0 'etrospective

• ongitudinal vs0 Cross-sectional

• 'andomized vs0 1on-'andomized

• /linded)Mas9ed or 1ot♦ Single-blind 2ouble blind =nblinded


61/98

'andomization! 2efinition

• 'andom Allocation♦ 9no*n c.ance receiving a treatment

♦ cannot predict t.e treatment to be given

• "liminate Selection /ias

• Similar Treatment Nroups


62/98

O1" %actor is 2ifferent

• 'andomization tries to ensure t.at O1"factor is different bet*een t*o or more

groups0• Observe t.e Conse?uences

• Attribute Causalit8


63/98

T8pes of 'andomization

• Standard *a8s!

♦ 'andom number tables Dsee te#tE

♦ Computer programs• 1OT legitimate

♦ /irt. date

♦

ast digit of t.e medical record number ♦ Odd)even room number


64/98

T8pes of 'andomization

• Simple

• /loc9ed 'andomization

• Stratified 'andomization


65/98

Simple 'andomization

• 'andomize eac. patient to a treatment*it. a 9no*n probabilit8

♦ Corresponds to flipping a coin• Could .ave imbalance in W ) group or

trends in group assignment

• Could .ave different distributions of atrait li9e gender in t.e t*o arms


66/98

/loc9 'andomization

• Insure t.e W of patients assigned toeac. treatment is not far out of balance

• (ariable bloc9 size♦ An additional la8er of blindness

• 2ifferent distributions of a trait li9e

gender in t.e t*o arms possible


67/98

Stratified 'andomization

• A priori certain factors li9el8 importantDe0g0 Age NenderE

• 'andomize so different levels of t.efactor are balanced bet*een treatmentgroups

• Cannot evaluate t.e stratificationvariable


68/98

Stratified 'andomization

• %or eac. subgroup or strata perform aseparate bloc9 randomization

• Common strata♦ Clinical center Age Nender

• Stratification M=ST be ta9en into

account in t.e data anal8sis;


69/98

Outline

• Introductor8 Statistical 2efinitions

• 7.at is 'andomization4

• 'andomized Stud8 2esign

• "#perimental vs0 Observational

• 1on-'andomized Stud8 2esign

• Stat Soft*are /oo9s Articles


70/98

T8pes of 'andomized Studies

• $arallel Nroup

• Se?uential Trials

• Nroup Se?uential trials

• Cross-over

• %actorial 2esigns


71/98

$arallel Nroup

• 'andomize patients to one of 9treatments

• 'esponse♦ Measure at end of stud8

♦ 2elta or R c.ange from baseline

♦ 'epeated measures♦ %unction of multiple measures


72/98

Ideal Stud8 - Nold Standard

• 2ouble blind

• 'andomized

• $arallel groups

T*o Scenarios


73/98

T*o Scenarios

• Stud8 J♦ A =0S0 stud8 D,FFFE compared LH patients *it. brain cancer

to L,, patients *.o did not .ave brain cancer0 T.e patients<cell p.one use *as measured using a ?uestionnaire0 T.e t*ogroups< use of cell p.ones *as similar0

• Stud8 ,♦ An Australian stud8 DJE conducted a stud8 *it. ,FF

transgenic mice0 One .undred *ere e#posed for t*o GFminute periods a da8 to t.e same 9ind of micro*aves *it.roug.l8 t.e same po*er as t.e 9ind transmitted from a cellp.one0 T.e ot.er JFF mice *ere not e#posed0 After J

mont.s t.e brain tumor rate for t.e e#posed mice *as t*iceas .ig. as t.at for t.e une#posed mice0


74/98

6uestions to Consider

• :o* do t.e t*o studies differ4♦ Stud8 J

♦ Stud8 ,


75/98


76/98

6uestions to Consider

• Suppose a friend recentl8 diagnosed *it.brain cancer *as a fre?uent cell p.one

user0 Is t.is strong evidence t.at fre?uentcell p.one use increases t.e li9eli.ood ofgetting brain cancer4♦ Informal observations of t.is t8pe are called

XXXXXXXXXXXXX XXXXXXXXXXXXX0♦ @ou s.ould rel8 on reputable researc. studies

not anecdotes0


77/98

T*o Main 7a8s to Nat.er 2ata

• Observational Stud8♦ T.e researc.er observes values of t.e response and

e#planator8 variables for t.e sampled sub>ects *it.outimposing an8 treatments

♦ "#ample!

• "#periment♦ T.e researc.er assigns e#perimental conditions Dalso

called treatmentsE to sub>ects Dalso called e#perimentalunitsE and t.en observes outcomes on t.e response

variable0♦ Treatments correspond to values of t.e e#planator8

variable♦ "#ample!


78/98

T8pes of Observational Studies

• 'etrospective♦ Observational studies t.at loo9 bac9 in time

• T.is is sometimes done to find ris9 factors for certain

diseases

• Cross-Sectional♦ Observational studies t.at ta9e a cross section of

t.e population at t.e current time

• $rospective♦ Observational studies in *.ic. sub>ects are

follo*ed into t.e future


79/98

Advantages of "#periments over

Observational Studies• In an observational stud8 t.ere can al*a8s be

lur9ing variables affecting t.e results0

• T.is means t.at observational studies can

XXXXXXXXX s.o* causation0

• It is easier to ad>ust for lur9ing variables in ane#periment0

• In general *e can stud8 t.e effect of an e#planator8variable on a response variable more accuratel8*it. an e#periment t.an *it. an observational stud80


80/98

2isadvantages of "#periments

• T.e8 can be XXXXXXXXXXXX to perform on t.esub>ects in *.ic. 8ou are interested0

• It can be difficult to monitor sub>ects to ensure t.att.e8 are doing *.at t.e8 are told0

• T.e8 can ta9e man8 8ears even decades tocomplete0

• 'esults of e#periments t.at use animals do not XXXXXXXXXXXXXX to .umans0

• T.e8 are unnecessar8 t.e ?uestion of interest doesnot involve tr8ing to assess XXXXXXXXXXXXX0


81/98

Sampling 2esigns for

Observational Studies• Simple 'andom Sampling DS'SE

♦ A simple random sample of n sub>ects from a population isone in *.ic. eac. possible sample of t.at size .as t.e

XXXXXXX c.ance of being selected0


82/98

Sampling 2esigns for ObservationalStudies

• Stratified Sampling♦ A stratified random sample divides t.e population into

separate groups called XXXXXXXX and t.en selects an S'S

from eac. stratum0


83/98


Observational Studies• Cluster Sampling

♦ A cluster random sample can be used if t.e target populationnaturall8 divides into groups eac. of *.ic. is representative of t.eentire target population0 In t.is met.od a S'S of groups Dor strataEis ta9en0 "ver8 member of t.e selected groups is put into t.e

sample0


84/98


Observational Studies• S8stematic Sampling♦ A s8stematic sample selects ever8 9t.

person from t.e sample frame0 T.eresearc.er randoml8 selects a numberbet*een J and 9 in order to 9no* *.ic.person to select first t.en selects ever8 9t.

person after t.is0

f


85/98

Advantages of t.e (arious

Sampling 2esigns• Simple 'andom Sampling DS'SE

♦ It is t.e easiest most *idespread form of sampling0

♦ "ac. sub>ect .as an XXXXXXX c.ance to be in t.esample0

♦ T.e sample enables us to determine .o* li9el8 it ist.at descriptive statistics Dli9e t.e sample meanEfall close to corresponding values for *.ic. *e*ould li9e to ma9e inference Dli9e t.e populationmeanE0

Ad f . ( i


86/98

Advantages of t.e (arious

Sampling 2esigns• Stratified Sampling

♦ It ensures t.at t.ere are enoug. XXXXXXXXX in

eac. group t.at 8ou *ant to compare0

• Cluster Sampling♦ It does not re?uire a sampling frame of sub>ects0

♦

It is less XXXXXXXXXXX to implement0


87/98

/ias in Sampling

• A sampling met.od is XXXXXXXXX if ♦ T.e sample tends to favor some parts of t.e

population over ot.ers0

♦ In ot.er *ords t.e results from t.e sample are notrepresentative of t.e population0

• Obviousl8 XXXXXXXXXX samples are our goal0


88/98

" l f $ S l t. t


89/98

"#amples of $oor Samples t.at

'esult in /ias• Convenience Samples

• (oluntar8 'esponse Samples


90/98

"lements of a Nood "#periment

• Control group♦ Nives us somet.ing to compare against♦ "nables us to control t.e XXXXXXXXXX XXXXXXX

• T.e placebo effect occurs *.en patients seem to improve

regardless of t.e treatment t.e8 receive0• 'andomization

♦ "liminates XXXXXX t.at can result *.en researc.ers assigntreatments to t.e sub>ects

♦ /alances t.e group on variables t.at 8ou 9no* affect t.e

response♦ /alances t.e group on XXXXXXXXX variables t.at ma8 beun9no*n to 8ou


91/98


92/98


93/98

"#ample

• To investigate *.et.er antidepressants .elp smo9ers to ?uit smo9ingone stud8 used L, men and *omen *.o *ere J or older and .adsmo9ed JK cigarettes or more per da8 in t.e previous 8ear0 T.e8 *ereall .ig.l8 motivated to ?uit and in good .ealt.0 T.e8 *ere assigned toone of t*o groups! one group too9 an antidepressant called Y8ban

*.ile t.e ot.er group did not ta9e an8t.ing0 At t.e end of a 8ear t.estud8 observed *.et.er eac. sub>ect .ad successfull8 abstained fromsmo9ing0


94/98


95/98

2id @ou Zno*4

• Observational studies can also .ave control groups0♦ T.ese are called ______-________ studies0♦ T.e cases are people *.o .ave a certain disease or

condition and t.e controls are people *.o do not .ave t.edisease0

♦ T.eir purpose is to see if one of t.e e#planator8 variables isrelated to t.e disease0

♦ XXXXXXXXX from t.e beginning of t.ese notes is an e#ampleof a case-control stud80


96/98

Important $oints

• T8pes of studies!♦ Observational studies and e#periments♦ "#periments control for lur9ing variables

• Sampling designs!♦

S'S stratified random samples and cluster samples♦ S'S is t.e preferred met.od• $otential sources of bias!

♦ =ndercoverage♦ 'esponse bias♦ 1onresponse bias♦

Convenience sampling♦ (oluntar8 response sampling

• "lements of good e#periments!♦ Control group randomization and blinding


97/98

Important $oints

• If a group is underrepresented in t.e sample *e cannotma9e inference about it0

• 7e must be careful *.en interpreting t.e results ofobservational studies0

• %or comparison of several treatments to be valid 8ou mustappl8 all treatments to similar groups of e#perimental units0

• Interesting ?uestions are usuall8 prett8 toug. to ans*er0 T.isis due in part to t.e fact t.at no single e#periment orobservational stud8 can determine causation0


98/98

Stop and *$in?;;;Stop and *$in?;;;

•Write the study!

•Describe & classify the

variables.

•Instruents for easure

•Bias

•"re#are to analy$e data!

Date post:	07-Jul-2018
Category:	Documents
Upload:	andreea-cambei
View:	226 times
Download:	0 times

Curs 3 CB Metodologia Cercetarii

Documents