HLTH2270 Lecture Notes - Lecture 5: Statistical Hypothesis Testing, Statistical Power, Histogram
Assumption Definition
a thig that is aepted as tue
o as etai to happe, ithout poof
From http://www.oxforddictionaries.com/definition/english/assumption
prejudgement, expectation,
hypothesis, suspicion
Statistical Assumptions
Many statistical test we use are based on assumptions we make about our data
If assumptions correct – strong statistical power!
If assumptions incorrect – results can be misleading!
Parametric v Non-Parametric Tests
• Types of statistical tests which adhere to specific criteria:
• Parametric tests require data to confirm to several assumptions:
1. Additivity and Linearity 4. Homogeneity of Variance
2. Interval or Ratio Data 5. Normal Distribution
3. Independence
• Non-Parametric tests are used when data violates the above assumptions
Parametric Tests- Additivity and Linearity
Arguably the most important!
Most tests use linear models to make inferences, therefore:
• We assume outcome variables are linearly related to any predictor
• If we have several predictors then their combined effect is best described by adding
their effects together.
Parametric Tests- Additivity and Linearity
• If this assumption is violated then your model will always be invalid
• If your data violates assumptions of additivity and linearity the model and parameter
estimates will be wrong
• It’s the euialet of allig ou at a dog:
• Just eause ou alled it a dog it o’t feth o sit he ou tell it to.
• It’s ehaious o’t hage o ake sese if ou itepeted the o the
assumption of it being a dog
find more resources at oneclass.com
find more resources at oneclass.com
Using graphs to spot linearity
We a plot stadadised esiduals…
It’s a easue of stegth etee oseed ad epeted alues
Convert predicted value and
error to z-scores
Parametric Tests- Interval Data
• Data should be either interval or ratio scale
• This is simply because it is impossible to have data normally distributed using any
other type of data or scale
Parametric Tests
Non-parametric Tests
Discrete
Nominal
Continuous
Ordinal
Binary
Parametric Tests- Independence
• Data from different participants are independent
i.e. Behaviour of one participant does not affect the behaviour of another
• Example of independence:
Without talking to aoe o itig athig do…. Reee the folloig ues i
order:
5; 7; 4; 8; 7; 1; 2; 3; 7; 4; 4; 6; 2; 1; 5
Now write the numbers down:
Cout ho a ou got i oet ode…
Do the sae agai… Hoee this time you can get the person to your right to help you:
6; 3; 2; 8; 7; 9; 8; 4; 7; 5; 1; 1; 3; 4; 7
No ite the ues do… You a disuss ith ou pate.
Cout ho a ou got i oet ode…
• Not only can dependency change outcome measures
• The equations used in statistical tests includes the standard error of the sample
• Standard Error estimates are only valid if the observations are independent
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
Assumption definition (cid:862)a thi(cid:374)g that is a(cid:272)(cid:272)epted as t(cid:396)ue o(cid:396) as (cid:272)e(cid:396)tai(cid:374) to happe(cid:374), (cid:449)ithout p(cid:396)oof(cid:863) Many statistical test we use are based on assumptions we make about our data. If assumptions incorrect results can be misleading! Most tests use linear models to make inferences, therefore: we assume outcome variables are linearly related to any predictor. If we have several predictors then their combined effect is best described by adding their effects together. If this assumption is violated then your model will always be invalid. If your data violates assumptions of additivity and linearity the model and parameter estimates will be wrong. It"s the e(cid:395)ui(cid:448)ale(cid:374)t of (cid:272)alli(cid:374)g (cid:455)ou(cid:396) (cid:272)at a dog: Just (cid:271)e(cid:272)ause (cid:455)ou (cid:272)alled it a dog it (cid:449)o(cid:374)"t fet(cid:272)h o(cid:396) sit (cid:449)he(cid:374) (cid:455)ou tell it to. It"s (cid:271)eha(cid:448)iou(cid:396)s (cid:449)o(cid:374)"t (cid:272)ha(cid:374)ge o(cid:396) (cid:373)ake se(cid:374)se if (cid:455)ou i(cid:374)te(cid:396)p(cid:396)eted the(cid:373) o(cid:374) the assumption of it being a dog. It"s a (cid:373)easu(cid:396)e of st(cid:396)e(cid:374)gth (cid:271)et(cid:449)ee(cid:374) o(cid:271)se(cid:396)(cid:448)ed a(cid:374)d e(cid:454)pe(cid:272)ted (cid:448)alues.
