postgraduate course evidence-based management (some) statistics for managers who hate statistics

62
Postgraduate Course Evidence-Based Management (Some) statistics for managers who hate statistics

Upload: melvyn-austin

Post on 23-Dec-2015

220 views

Category:

Documents


0 download

TRANSCRIPT

Postgraduate Course

Evidence-Based Management

(Some) statistics for managers who hate statistics

Postgraduate Course

Why do we need statistics?

1. How does my population look like?

2. Is there a difference?

3. Is there a model that ‘fits’?

Postgraduate Course

Some statistics

Some statistic terms

1. Sample vs population

2. Variables

3. Levels of measurement

4. Central tendency

5. Hypothesis

Some statistic models

6. Mean

7. Variance, standard deviation

8. Confidence intervals

9. Statistical significance

10. Statistical power

11. Effect sizes 12.Critical appraisal

Postgraduate Course

1. Sample vs population

Postgraduate Course

Sample vs population

We want to know about these(population: N)

We have to work with these(sample: n)

population mean: μ

selection

sample mean: X _

statistics

fit?

Postgraduate Course

Law of large numbers

The larger the sample size (or the number of

observations), the more accurate the predictions of the

characteristics of the whole population, and smaller

the expected deviation in comparisons of outcomes.

As a general principle it means that, in the long run,

the average (mean) of a large number of observations

will be close to (or: may be taken as the best estimate

of) the 'true mean’ of the population.

Sample vs population

Postgraduate Course

Sample size: why does it matter?

Law of the large numbers: a reliable and accurate

representation of the population

Statistical power: to prevent a type 2 error / false

negative

Sample vs population

Don’t confuse: representativeness and reliability

The sample size has no direct relationship with

representativeness; even a large random sample can be

insufficiently representative.

Postgraduate Course

Sample vs population

Postgraduate Course

2. VariablesPostgraduate Course

Postgraduate Course

VariablesPostgraduate Course

Variable: anything that can be measured and can differ across entities or time

Independent variable: predictor variable (value does not depend on any other variables)

Dependent variable: outcome variable (value depends on other variables)

Postgraduate Course

3. Level of measurementPostgraduate Course

Postgraduate Course

Level of measurementPostgraduate Course

Relationship between what is being measured and the numbers that represent what is being measured.

Postgraduate Course

Categorical

Continuous

Nominal

Ordinal

Interval

Ratio

Level of measurement

Postgraduate Course

Nominal scale

Classification of categorical data. There is no order to the values, they are just given a name (‘nomen’) or a number. The numbers can’t be used to calculate … (you can’t calculate the mean of fruit) .. only frequencies

1 = Apples2 = Oranges3 = Pineapples4 = Banana’s5 = Pears6 = Mango’s

Postgraduate Course

Ordinal scale

Classification of categorical data. Values can be

rank-ordered, but the distance between the

values have no meaning. The numbers can only

be used to calculate a modus or a median

1. Full Professor2. Associate professor3. Assistant professor4. PhD5. Master6. Bachelor

Postgraduate Course

Interval scale

Classification of continuous data. Values can

be rank-ordered, and the distance between

the values have meaning. However, there is

no natural zero point

1. John (1932)2. Denise (1945)3. Mary (19524. Marc (1964)5. Jeffrey (1978)6. Sarah (1982)

Postgraduate Course

Ratio scale

Classification of continuous data. Values can

be rank-ordered, the distance between the

values have meaning and there is a natural

zero point.

1. Jeffrey (192 cm)2. John (187 cm)3. Sarah (180 cm4. Marc (179 cm)5. Mary (171 cm)6. Denise (165 cm)

Postgraduate Course

Nominal Ordinal Interval Ratio

Classification Yes Yes Yes Yes

Rank-order No Yes Yes Yes

Fixed and equal intervals No No Yes Yes

Natural 0 point No No No Yes

Nominal Ordinal Interval Ratio

Mode Yes Yes Yes Yes

Median No Yes Yes Yes

Mean No No Yes Yes

Levels of measurement

Categorical Continuous

Postgraduate Course

Levels of measurement

Ordinal or interval? Can I calculate a mean?

Q3: Every organization is unique, hence the findings from scientific research are not applicable.

☐ Strongly agree

☐ Somewhat agree

☐ Neither agree or disagree

☐ Somewhat disagree

☐ Strongly disagree

Postgraduate Course

4. Central tendency

The aim is to find a single number that characterises the typical value of the variable in the sample. Which one you use depends in part on the level of measurement of the variable.

Postgraduate Course

Central tendency

Central tendency of a set of data / numbers

(what number is most representative of the dataset / population?)

7, 9, 9, 9, 10, 11,11, 13, 13

Mean = 10,2

Median = 10

Mode = 9

Postgraduate Course

Central tendency

Central tendency of a set of data / numbers

(what number is most representative of the dataset / population?)

3, 3, 3, 3, 3, 3, 100

Mean = 16,9

Median = 3

Mode = 3

Postgraduate Course

5. Hypothesis

Postgraduate Course

“It is easy to obtain evidence in favor of virtually any theory,

but such ‘corroboration’ should count scientifically only if it

is the positive result of a genuinely ‘risky’ prediction, which

might conceivably have been false.

… A theory is scientific only if it is refutable

by a conceivable event. Every genuine test

of a scientific theory, then, is logically an

attempt to refute or to falsify it.”

Hypothesis: falsifiability

Carl Popper

Postgraduate Course

Hypothesis

Null hypothesis (H0): Big Brother contestants and

members of the public will not differ in their scores on

personality disorder questionnaires

Alternative hypothesis (H1): Big Brother contestants will

score higher on personality disorder questionnaires

than members of the public.

Postgraduate Course

Hypothesis: type I vs type II error

null hypothesis is true

& was rejected(type I error)

α

null hypothesis is false

& was rejected(correct conclusion)

null hypothesis is true

& was accepted(correct conclusion)

null hypothesis is false

& was accepted(type II error)

β

H0 is true H0 is false

reject H0

accept H0

Postgraduate Course

Statistic models

Postgraduate Course

Statistic models: prediction

likely not likely

Postgraduate Course

6. The mean

The most widely used statistic model

μX_

or

sample population

Postgraduate Course

The mean

EBMgt Lecturer

Num

ber o

f Frie

nds

Postgraduate Course

The mean

Assessing the fit of the mean

Sum of squared errors (SS): (-1,6) + (-0,6) + (0,4) + (0,4) + (1,4) = 5,2

Variance (s ): = = 1,3

Standard deviation (s): √s = 1,14

2 2 2 2 2

2 SSN-1

5,24

2

Postgraduate Course

The second most widely used statistic model

σs or

sample population

7. Standard Deviation

Postgraduate Course

Standard Deviation

Postgraduate Course

110IQ

Postgraduate Course

Standard Deviation

Which class would you prefer to teach?

130 170

Postgraduate Course

110 130IQ

S=10

S=20

S=60

170

Postgraduate Course

Standard Deviation

Postgraduate CoursePostgraduate Course

Standard Deviation

Postgraduate CoursePostgraduate Course

So, what does

“two standard deviations of the mean”

mean?

Standard Deviation

Postgraduate Course

8. Confidence intervalsPostgraduate Course

Postgraduate Course

A confidence interval gives an estimated range

of values which is likely to include an unknown

population parameter (e.g. the mean).

Confidence intervals are usually calculated so

that this percentage is 95% (95% CI)

Confidence intervals

Postgraduate Course

When you see a 95% confidence interval for a

mean, think of it like this: if we’d collected 100

samples and calculated the mean for each

sample, than for 95 of these samples the mean

would fall within the confidence interval.

Confidence intervals

Postgraduate Course

1,96!

Confidence intervals

Postgraduate Course

Confidence intervals

Postgraduate Course

2008 2009

4,5

4,0

3,5

5,0

3,0

“According to the federal

government, the

unemployment rate has

dropped from 4.3% to 3.8%.”

95% CI= 4,1 - 3,5.

This means the

unemployment rate could

have increased from 4.0 to

4,1 !

Confidence intervals

Postgraduate Course

When a point estimate (e.g. mean,

percentage) is given, always check:

standard deviation

or

confidence interval

Confidence intervals

Postgraduate Course

9. Statistical significance

Postgraduate Course

Statistical significance

Sir Ronald A. Fisher1890 - 1962

Significant = the probability of incorrectly rejecting the

null hypothesis (= Type I error, α)

p = 0,05 / p = 0,01

Postgraduate Course

Statistical significance

(1 in 20 / 1 in 100)

Postgraduate Course

Statistical significance

110 130

Postgraduate Course

1. Is there a difference / an effect?

2. How certain is it that the difference / effect found is not a

chance finding?

X_

0 X_

1

Statistical significance

Testing multiple hypothesis

When you test 20 different hypotheses (or independent

variables), there is a high chance that at least one will be

statistically significant.

example:

Does apples, bacon, cheese, eggs, fish, garlic, hazelnuts, ice

cream, ketchup, lamb, melons, nuts, oranges, peanut butter,

roasted food, salt, tofu, vinegar, wine or yoghurt cause

cancer?

Postgraduate Course

Statistical significance

Significance testing:

always prospective, never retrospective

Postgraduate Course

Statistical significance

Statistical significant ≠ practical relevant

Postgraduate Course

Effect size

Statistical significance

Postgraduate Course

10. Statistical power

Sample size Effect size (Significant increase in IQ)

4 10

25 4

100 2

10.000 0,2

Postgraduate Course

Statistical power

The statistical power: the power to detect a meaningful

effect, given sample size, significance level, and effect size.

Postgraduate Course

Overpowered: sample size too large, high

probability of making a Type I error

Underpowered: sample size too small, high

probability of making a Type II error.

Statistical power

Postgraduate Course

11. Effect size

Postgraduate Course

Effect size

Effect size: a standardized measure of the

magnitude of effect, independent of

sample size

standardized > makes it possible to compare effect sizes

across different studies that have measured different

variables, or have used different scales of measurement

Postgraduate Course

Effect sizes

Cohen’s d

Pearson’s r

other - Hedges’ g

- Glass’ Δ

- odds ratio OR

- relative risk RR

Postgraduate Course

Effect sizes

Cohen’s d

Effect size based on means or distances

between/among means

Interpretation

< .10 = small

.30 = moderate

> .50 = large

Postgraduate Course

Effect sizes

Pearson’s r

Effect size based on ‘variance explained’

Interpretation

< .10 = small (explains 1% of the total variance)

.30 = moderate (explains 9% of the total variance)

> .50 = large (explains 25% of the total variance)

Postgraduate Course

12. Critical appraisal

When you critically appraise a study, what characteristics

of the findings will you consider to determine its statistical

significance and magnitude?

Postgraduate Course

Critical appraisal

When you critically appraise a study, what characteristics

of the findings will you consider to determine its statistical

significance and magnitude?

p-value

confidence interval

sample size / power

effect size

practical relevance