08/10/2013

Issues in experimental
design for the study of
atypical language
development
David Saldaña

DESIGNING YOUR STUDY: CONTROL

1
08/10/2013

Controlled experiments
This type of experiment is conducted in a well-controlled environment –
not necessarily a laboratory – and therefore accurate measurements are
possible.

Easier to replicate
Precise control of
extraneous and
independent
variables.

The artificiality of
the setting and lack
of generalization
Demand
characteristics or
experimenter effects
may bias the results
McLeod, S. A. (2012).

Field experiments
The experimenter still manipulates the independent variable, but in a reallife setting (so cannot really control extraneous variables).

More likely to reflect real life
because of it natural setting

Less control over extraneous
variables

Less likelihood of demand
characteristics affecting the
results

Less replicable

Can be used in situations in
which it would be ethically
unacceptable to manipulate
the independent variable, e.g.
researching stress.
McLeod, S. A. (2012).

2
08/10/2013

Natural experiments
Natural Experiments are conducted in the everyday environment of the participants
but here the experimenter has no control over the IV as it occurs naturally in real
life.

More likely to reflect real life because No control over extraneous variables
of it natural setting
Less replicable
Less likelihood of demand
Possible more expensive
characteristics affecting the results
Can be used in situations in which it
would be ethically unacceptable to
manipulate the independent variable,
e.g. researching stress.
Less likelihood of demand
characteristics affecting the results,
as participants may not know they
are being studied.

McLeod, S. A. (2012).

A perhaps “silly” research question

Does vocabulary
influence the
comprehension of
questions more than the
comprehension of
imperatives?

3
08/10/2013

A question that is a little more complicated

Do children with SLI
understand questions
better than typically
developing children?
Do children with SLI
understand questions
better than imperatives?

DESIGNING YOUR STUDY:
CONDITIONS

4
08/10/2013

How is your IV distributed?

Independent Measures

Repeated Measures
Matched Pairs

Independent measures (between-group)

• Each group gets one condition
• Different participants in each
group
• Avoids practice
• More people needed
• Participant variables could affect
results.

5
08/10/2013

Repeated measures (within-group)
Task 1

Task 2

• Measurement

• Measurement

- Less participants
- Precision determined by variation within same
subject
- May be the only design that answers the
questions of interest.
- Order effects: practice and fatigue effect

Counterbalancing

Moment 1

Moment 2

A

B

B

A

6
08/10/2013

Back to our “silly” study
Lets design the tasks and
the data collection

Does vocabulary
influence the
comprehension of
questions more than the
comprehension of
imperatives?

A little bit more interesting (perhaps)
research question

Does vocabulary
influence the
comprehension of
questions more than the
comprehension of
imperatives in SLI?

7
08/10/2013

Mixed-design
Group 1

Task 1

Task 2

• Measurement

• Measurement

Group 2

Task 1

Task 2

• Measurement

• Measurement

Some imaginative results
SLI

Typical

120

100

Standard Scores

100

99

95

92

90

80
77
60

70

73

90

90

90

75

40

20

0
IQ

Vocabulary

Comprehension
Questions

Word Reading

SES

Comprehension
Imperatives

8
08/10/2013

MATCHING

Matching your groups: why?
SLI

Typical

120

100

Standard Scores

100

99

95

92

90

80
77
60

70

73

90

90

90

75

40

20

0
IQ

Vocabulary

Comprehension
Questions

Word Reading

SES

Comprehension
Imperatives

9
08/10/2013

Matching your groups: on what?
6

8 10

2

4

6

8

2

4

6

8

Imp.Comp

6

8 10

IQ

8

5

10 15

QuestComp

4

6

Voc

8 10

2

4

6

SES

2

4

6

8

5

10 15

4

6

8 10

Your groups: three group design

• Group of interest

• Control group, matched on chronological age

• Control group, matched on variable of interest

10
08/10/2013

Some issues
• Match on variables of interest to
the study
• Match average and distribution
• Study carefully your exclusions

Recruit clinical
sample

Decide
variables of
matching

Recruit age
controls

Recruit “level”
controls

Let’s do it! Matching exercise

Some issues
Recruit clinical
sample

• Match on variables of interest to
the study
• Match average and distribution
• Study carefully your exclusions

Decide
variables of
matching

Recruit age
controls

Recruit “level”
controls

Check and
report

11
08/10/2013

What statistical test?
An ideal world

Facon et al. (2011)

What statistical test?
An ideal world

12
08/10/2013

What statistical test?
An ideal world

The real world

Non-normal distributions
Different range and variance
Different distributions

What statistical test?
None: look at the picture first!

Back-to-back histogram

Back-to-back stem-and-leaf

13
08/10/2013

What statistical test?

What statistical test?
Mann-Whitney?
Two-sample Kolmogorov-Smirnov test
For small sample sizes: exact version
• Does not assume prior “shape” in distributions
• Tests for the differences in distribution

14
08/10/2013

Matching your groups: how much?
The easy-peasy way

Matching your groups: how much?
A brief recall of significance testing

I say
No Match
I say
Match
Match

No Match

http://intuitor.com/statistics/T1T2Errors.html

15
08/10/2013

“Casual acceptance of the null hypothesis”
(Harcum, 1990)

< .2 too low
Frick (1995)

.2 to .5 ambiguous
> .5 fine

http://intuitor.com/statistics/T1T2Errors.html

Some things to remember
• Population comparisons are group comparisons

(Paradis, 2010)

16
08/10/2013

Some things to remember
• Each group must be there for a reason

(Paradis, 2010)

An example: why each group?
Two groups of deaf adolescents
SAL

16 deaf (13-21) good oral language

SBL

16 deaf (13-21) poor oral language

Three groups of hearing adolescents
OCS

20 adolescents (13-21) task with sound

OSS

20 adolescents (13-21) task with no sound

NO

20 children (6-11) oral language equivalent to SAL
Torres (2013)

17
08/10/2013

Some things to remember
• Check matching while you are recruiting

(Paradis, 2010)

Keep your lab notes up to date
•Some important
decisions to record:
•Changes and
discussions on criteria
•Subjects in – subjects
out.
•Transcriptions and
criteria if using MLU
(Paradis, 2010)

18

Issues in experimental design for the study of atypical language development

  • 1.
    08/10/2013 Issues in experimental designfor the study of atypical language development David Saldaña DESIGNING YOUR STUDY: CONTROL 1
  • 2.
    08/10/2013 Controlled experiments This typeof experiment is conducted in a well-controlled environment – not necessarily a laboratory – and therefore accurate measurements are possible. Easier to replicate Precise control of extraneous and independent variables. The artificiality of the setting and lack of generalization Demand characteristics or experimenter effects may bias the results McLeod, S. A. (2012). Field experiments The experimenter still manipulates the independent variable, but in a reallife setting (so cannot really control extraneous variables). More likely to reflect real life because of it natural setting Less control over extraneous variables Less likelihood of demand characteristics affecting the results Less replicable Can be used in situations in which it would be ethically unacceptable to manipulate the independent variable, e.g. researching stress. McLeod, S. A. (2012). 2
  • 3.
    08/10/2013 Natural experiments Natural Experimentsare conducted in the everyday environment of the participants but here the experimenter has no control over the IV as it occurs naturally in real life. More likely to reflect real life because No control over extraneous variables of it natural setting Less replicable Less likelihood of demand Possible more expensive characteristics affecting the results Can be used in situations in which it would be ethically unacceptable to manipulate the independent variable, e.g. researching stress. Less likelihood of demand characteristics affecting the results, as participants may not know they are being studied. McLeod, S. A. (2012). A perhaps “silly” research question Does vocabulary influence the comprehension of questions more than the comprehension of imperatives? 3
  • 4.
    08/10/2013 A question thatis a little more complicated Do children with SLI understand questions better than typically developing children? Do children with SLI understand questions better than imperatives? DESIGNING YOUR STUDY: CONDITIONS 4
  • 5.
    08/10/2013 How is yourIV distributed? Independent Measures Repeated Measures Matched Pairs Independent measures (between-group) • Each group gets one condition • Different participants in each group • Avoids practice • More people needed • Participant variables could affect results. 5
  • 6.
    08/10/2013 Repeated measures (within-group) Task1 Task 2 • Measurement • Measurement - Less participants - Precision determined by variation within same subject - May be the only design that answers the questions of interest. - Order effects: practice and fatigue effect Counterbalancing Moment 1 Moment 2 A B B A 6
  • 7.
    08/10/2013 Back to our“silly” study Lets design the tasks and the data collection Does vocabulary influence the comprehension of questions more than the comprehension of imperatives? A little bit more interesting (perhaps) research question Does vocabulary influence the comprehension of questions more than the comprehension of imperatives in SLI? 7
  • 8.
    08/10/2013 Mixed-design Group 1 Task 1 Task2 • Measurement • Measurement Group 2 Task 1 Task 2 • Measurement • Measurement Some imaginative results SLI Typical 120 100 Standard Scores 100 99 95 92 90 80 77 60 70 73 90 90 90 75 40 20 0 IQ Vocabulary Comprehension Questions Word Reading SES Comprehension Imperatives 8
  • 9.
    08/10/2013 MATCHING Matching your groups:why? SLI Typical 120 100 Standard Scores 100 99 95 92 90 80 77 60 70 73 90 90 90 75 40 20 0 IQ Vocabulary Comprehension Questions Word Reading SES Comprehension Imperatives 9
  • 10.
    08/10/2013 Matching your groups:on what? 6 8 10 2 4 6 8 2 4 6 8 Imp.Comp 6 8 10 IQ 8 5 10 15 QuestComp 4 6 Voc 8 10 2 4 6 SES 2 4 6 8 5 10 15 4 6 8 10 Your groups: three group design • Group of interest • Control group, matched on chronological age • Control group, matched on variable of interest 10
  • 11.
    08/10/2013 Some issues • Matchon variables of interest to the study • Match average and distribution • Study carefully your exclusions Recruit clinical sample Decide variables of matching Recruit age controls Recruit “level” controls Let’s do it! Matching exercise Some issues Recruit clinical sample • Match on variables of interest to the study • Match average and distribution • Study carefully your exclusions Decide variables of matching Recruit age controls Recruit “level” controls Check and report 11
  • 12.
    08/10/2013 What statistical test? Anideal world Facon et al. (2011) What statistical test? An ideal world 12
  • 13.
    08/10/2013 What statistical test? Anideal world The real world Non-normal distributions Different range and variance Different distributions What statistical test? None: look at the picture first! Back-to-back histogram Back-to-back stem-and-leaf 13
  • 14.
    08/10/2013 What statistical test? Whatstatistical test? Mann-Whitney? Two-sample Kolmogorov-Smirnov test For small sample sizes: exact version • Does not assume prior “shape” in distributions • Tests for the differences in distribution 14
  • 15.
    08/10/2013 Matching your groups:how much? The easy-peasy way Matching your groups: how much? A brief recall of significance testing I say No Match I say Match Match No Match http://intuitor.com/statistics/T1T2Errors.html 15
  • 16.
    08/10/2013 “Casual acceptance ofthe null hypothesis” (Harcum, 1990) < .2 too low Frick (1995) .2 to .5 ambiguous > .5 fine http://intuitor.com/statistics/T1T2Errors.html Some things to remember • Population comparisons are group comparisons (Paradis, 2010) 16
  • 17.
    08/10/2013 Some things toremember • Each group must be there for a reason (Paradis, 2010) An example: why each group? Two groups of deaf adolescents SAL 16 deaf (13-21) good oral language SBL 16 deaf (13-21) poor oral language Three groups of hearing adolescents OCS 20 adolescents (13-21) task with sound OSS 20 adolescents (13-21) task with no sound NO 20 children (6-11) oral language equivalent to SAL Torres (2013) 17
  • 18.
    08/10/2013 Some things toremember • Check matching while you are recruiting (Paradis, 2010) Keep your lab notes up to date •Some important decisions to record: •Changes and discussions on criteria •Subjects in – subjects out. •Transcriptions and criteria if using MLU (Paradis, 2010) 18