Stats without Tears
Seven Steps of Hypothesis Tests
Updated 3 Nov 2020
(What’s New?)
Copyright © 2010–2023 by Stan Brown, BrownMath.com
Updated 3 Nov 2020
(What’s New?)
Copyright © 2010–2023 by Stan Brown, BrownMath.com
Advice: Always number your steps. That helps others find the key features of your test, and you don’t forget any steps.
See also:
Inferential Statistics: Basic Cases
Top 10 Mistakes of Hypothesis Tests
Following are patterns for your hypotheses in the cases covered in the text. With Cases 1 through 5, if you can say anything meaningful about the consequences if each hypothesis is true, add that.
Bad example (adds little or nothing to the symbols):
H0: μ = 67.6, average 2-liter bottle contains 67.6 fl oz
H1: μ < 67.6, average 2-liter bottle contains less than 67.6 fl oz
Good example (explains the implications):
H0: μ = 67.6, average bottle filled properly
H1: μ < 67.6, average bottle is underfilled
In Cases 1 through 5, a test for < or > is called a one-tailed test, and a test for ≠ is called a two-tailed test. Please see One-Tailed or Two-Tailed? for advice on choosing between them.
Case 1:
(Testing mean of one population against a number
called μo)
H0: μ = number
H1: μ < number or μ ≠ number or μ > number
Case 2:
(Testing proportion in one population
against a number called po)
H0: p = number
H1: p < number or p ≠ number or p > number
Case 3:
(Testing mean difference (paired data))
d = _____ − _____
H0: μd = 0
H1: μd < 0 or μd ≠ 0 or μd > 0
Case 4:
(Testing difference of independent means)
pop. 1 = _____, pop. 2 = _____
H0: μ1 = μ2
H1: μ1 < μ2 or μ1 ≠ μ2 or μ1 > μ2
Case 5:
(Testing difference of population proportions)
pop. 1 = _____, pop. 2 = _____
H0: p1 = p2
H1: p1 < p2 or p1 ≠ p2 or p1 > p2
Case 6:
(Testing goodness of fit)
H0: The _____ model is consistent with the data.
H1: The model is not consistent with the data.
Case 7:
(Testing independence)
H0: _____ and _____ are independent.
H1: _____ and _____ are dependent.
Case 7:
(Testing homogeneity)
H0: The proportions are all equal.
H1: Some proportions are different from others.
Short and sweet:
α = _____
Please see Inferential Statistics: Basic Cases for specific requirements. For Cases 6 and 7, it’s easier to check requirements if you move this step after Steps 3/4.
Show screen name. Example: T-Test
. You
don’t need to write down keystrokes, such as “STAT TESTS
2”.
Show all inputs.
Show new outputs, meaning any that weren’t on the input screen.
No room for creativity here. Write down whichever one of these applies:
p < α. Reject H0 and accept H1.
p > α. Fail to reject H0.
Here you have a lot of latitude as long as you state the correct conclusion in English and give the significance level or p-value, or both.
If you rejected H0, state H1 without doubting words like may or could. Examples:
At the 0.05 significance level, the average 2-liter bottle contains less than 67.6 fl oz. Drinkems is underfilling the bottles.
Or,
The average 2-liter bottle contains less than 67.6 fl oz. Drinkems is underfilling the bottles (p = 0.0246).
If you failed to reject H0, state your non-conclusion in neutral language, using phrases like can’t determine whether or it’s impossible to say whether. Examples:
At the 0.05 significance level, we can’t tell whether Drinkems is underfilling the bottles or not.
Or,
We can’t tell whether Drinkems is underfilling the bottles or not (p = 0.1045).
Updates and new info: https://BrownMath.com/swt/