It Is Hard To Explain That Which Is (Mostly) Unexplained

12
Why Specific Cognitive Processing Weaknesses Are Typically Only Partial Explanations for Academic Deficits: It Is Hard To Explain That Which Is (Mostly) Unexplained W. Joel Schneider Department of Psychology

description

Why Specific Cognitive Processing Weaknesses Are Typically Only Partial Explanations for Academic Deficits:. It Is Hard To Explain That Which Is (Mostly) Unexplained. W . Joel Schneider Department of Psychology. Assumptions. Perfect multivariate normality Linear relationships only - PowerPoint PPT Presentation

Transcript of It Is Hard To Explain That Which Is (Mostly) Unexplained

Page 1: It Is Hard To Explain That Which Is (Mostly) Unexplained

Why Specific Cognitive Processing Weaknesses Are Typically Only Partial

Explanations for Academic Deficits:It Is Hard To Explain

That Which Is (Mostly) Unexplained

W. Joel SchneiderDepartment of Psychology

Page 2: It Is Hard To Explain That Which Is (Mostly) Unexplained

Assumptions• Perfect multivariate normality• Linear relationships only• Correlations are population parameters.

These assumptions are unlikely to be strictly true but they are

probably not that far from reality.

Page 3: It Is Hard To Explain That Which Is (Mostly) Unexplained

Difference = 0

Difference = 25

Difference = 25

Cognitive Strength

Cognitive Weakness

Academic Deficit

Gc = 100

Reading Decoding =75

Phonological Awareness =75

Gc = 100

Phonological Awareness =75

Does this weakness explain the academic deficit?

Page 4: It Is Hard To Explain That Which Is (Mostly) Unexplained

rGc.PA0.68

0.420

0.264

RD = b0 + b1 * Gc + b2 * PA + error

75 = 31.55 + 0.420 * 100 + 0.264 * 75 + -18.4

Ŷ = 93.4Predicted RD when

Gc = 100 and PA = 75

Error~N(0,11.642)

R2 = 0.40

Gc~N(100,152)

PA~N(100,152)

RD~N(100,152)

1

Ŷ ~N(100,9.462)

1

31.55 1.00

RD Gc PA

RD 1 0.60 0.54

Gc 0.60 1 0.68

PA 0.54 0.68 1

b1

b2

b0

RD = b0 + b1Gc + b2PA + error

ErrorError = -18.4

Ŷ = 93.4Ŷ

Ŷ ~N(100,(15R)2)

Error ~N(0,152(1 - R2))

Reading Decoding =75

Gc = 100

Phonological Awareness =75

Multiple R = rŶ.RD

Multiple Regression

Correlationsfrom WJ III NU (ages 9 to 13):

Page 5: It Is Hard To Explain That Which Is (Mostly) Unexplained

40 50 60 70 80 90 100 110 120 130 140 150 160

Reading Decoding(Gc = 100, PA = 75)

~N(93.4,11.642)

Reading Decoding(Whole Population)

~N(100,152)

Width of 95% CI = 59Width of 95% CI = 46

This is what “40% of the variance explained” looks like.

So, instead of a rangeof nearly 4 SDs, the

95% CI has narrowedto “only” 3 SDs.

Reading Decoding = 75 is somewhat unusual in both distributions.

6.6 points ≈ 0.5 SD

Page 6: It Is Hard To Explain That Which Is (Mostly) Unexplained

40

50

60

70

80

90

100

110

120

130

140

150

160

40 50 60 70 80 90 100 110 120 130 140 150 160

Reading Decoding

Predicted Reading Decoding (Ŷ)

RD distribution whenGc = 100 and PA = 75

~N(93.4,11.642)

5.7% of kids with Gc = 100 and PA = 75 have RD ≤ 75

Predicted RD = 93.4 when Gc = 100 and PA = 75

Page 7: It Is Hard To Explain That Which Is (Mostly) Unexplained

40

50

60

70

80

90

100

110

120

130

140

150

160

40 50 60 70 80 90 100 110 120 130 140 150 160

Reading Decoding

Predicted Reading Decoding (Ŷ)

RD distribution whenGc = 100 and PA = 100

~N(100,11.642)

1.6% of kids with Gc = 100 and PA = 100 have RD ≤ 75

Predicted RD = 100 when Gc = 100 and PA = 100

5.7 %1.6 % ≈3.6 ×  risk   of  𝑅𝐷≤75

Gc = 100PA = 100

Gc = 100PA = 75

Page 8: It Is Hard To Explain That Which Is (Mostly) Unexplained

The risk of low performance is low for people with either profile, but

the relative risk of low performance is much higher when PA is low. The

relative risk of low performance increases if the threshold for

defining “low performance” is lower.

Page 9: It Is Hard To Explain That Which Is (Mostly) Unexplained

RD distribution whenGc = 100 and PA = 6

~N(75,11.642)

50% of kids with Gc = 100 and PA = 6 have RD ≤ 75

Predicted RD = 75 when Gc = 100 and PA = 6

PA must be less than 6 before it is typical for a person with Gc = 100 to

have RD ≤ 75.

In a normal distribution, only about 1.2 people in the whole world have a

score 6 or lower!

If Gc = 100, how low does PA have to be

before we have a good explanation of RD = 75?

Page 10: It Is Hard To Explain That Which Is (Mostly) Unexplained

What about Multiple Deficits?

Gc Gf Gv Ga Glr Gsm Gs70

75

80

85

90

95

100

105

No Deficits 4 Processing DeficitsRD~N(100,11.242)1.3% risk of RD≤75

RD~N(90.5,11.242)8.4% risk of RD≤75

Relative Risk = = 6.4

Page 11: It Is Hard To Explain That Which Is (Mostly) Unexplained

ConclusionsWe must abandon dichotomous thinking. The question is not whether some ability is relevant to some outcome, but how much. Often, as in the demonstration here, a relevant predictor is merely a risk factor for low performance and is unlikely to be a sufficient explanation.

Page 12: It Is Hard To Explain That Which Is (Mostly) Unexplained

Current Context

Best Available Data and Models