Вы находитесь на странице: 1из 4

Key – Practice Exam 1

1) a) X ij = µ i + ε ij where µ i = average yield of variety i where I = 5.


Distribution assumption: ε ij ~ N (0, σ 2 ) and the ε ij are mutually independent.

b) Ho: µ A = µ B = µ C = µ D = µ E versus HA: at least 2 means differ

c) MSTrt = SSTrt/(I-1) = (SST – SSE)/(5-1) = (2609.04 – 1544.000)/4 = 266.26

266.26
d) F= = 3.449
77.2
e) F.05, 4, 20 = 2.87 and F.01, 4, 20 = 4.43 so the p-value is between .010 and .050.

f) α = .05 and the p-value < α so we should reject the null hypothesis.

Conclusion: The data provides statistical evidence (p-value between .01 and .05) that the average yield
differs between at least 2 varieties.

g) The experiment-wise-error rate is the probability of rejecting at least one of K true null hypothesis
in a set of K hypothesis tests done at the same time.

h) The value of 0 isn’t in the confidence interval for µ A − µ C but is in the Tukey interval because
Tukey intervals are calculated using Q which is larger than the t critical value. Therefore, the Tukey
intervals are wider than confidence intervals when the standard errors are approximately equal.

i) We use Tukey’s HSD Multiple Comparisons Test because it controls the experiment-wise-error
rate to be .05. If we did 10 separate Independent Samples T tests, the chance of making at least one
Type I error would be greater than .05.

2) SSTrt describes the variability between children’s HLT level that can be explained by differences
in the treatment method they received

• SSE describes the unexplained variability between children’s HLT level for children receiving the
same treatment method

• SST describes the total variability in children’s HLT levels

• MSE = SSE / I ( J − 1) = 1047.1/3(8-1) = 1047.1 / 21 = 49.862

• MSTrt = SSTrt / ( I − 1) = 309.2/2= 154.6


MSTrt 154.6
• The TS is F = = = 3.10
MSE 49.86

• The rejection region is : Reject the null hypothesis if TS ≥ 3.47 = F.05,2,21. Because the
TS < 3.47, we fail to reject the null hypothesis.

• The data does not provide statistical evidence that children’s HLT levels are affected differently
by the different methods of reducing hostility.

1
3) a) Model: X ijk = µ + α i + β j + γ ij + ε ijk where ε ijk ~ N (0, σ 2 ) , I = 3, J = 3 and K ij = 6
X ijk = shipping time for the kth package that was shipped at time of day i by firm j
µ = average shipping time averaged over all times of day and all firms
α i = the effect on shipping time due to level i time of day
β j = the effect on shipping time due to firm j
γ ij = the effect on the shipping time due to the interaction between level i time of day and
level j firm
ε ijk = random amount the observed shipping time varies from the expected shipping time

b) First check if there is an interaction (if there is, we don’t need to check the other rejection regions).

Interaction hypotheses: HoAB: γ ij =0 for all i, j vs HaAB: at least two γ ij j ≠ 0


Interaction hypotheses test whether the effect of time of day interacts with Firm to affect shipping
time.

SSAB / ( I − 1)( J − 1) MSAB 66.074 / 4


TS = TS = = = = 5.186
SSE / IJ ( K − 1) MSE 143.333 / 45
The rejection region is: Reject HoAB if TS ≥ F.05,4,45 which is between 2.61 and 2.56. The TS is in the
rejection region so the researcher should reject the “no interaction” hypothesis.

Conclusion: There is statistical evidence that the effect on shipping of time of day depends on the
firm.

My remark: When we conclude there is an interaction, we don’t make any conclusions about the
main effects.

4) a) ANS: HoA: α1 = α2 = α3 =0 vs HaA: at least two αI ≠ 0


Hypotheses for testing whether the of type of coat has an effect on body temperature

HoB: β1 = β2 = 0 vs HaB: β1 ≠ β2 ≠ 0
Hypotheses for testing whether the of strain of cattle has an effect on body temperature

HoAB: γ ij =0 for all i, j vs HaAB: at least two γ ij j ≠ 0


Hypotheses for testing whether effect of type of coat on body temperature depends upon the of
strain of cattle

b) There is no statistical evidence for an interaction effect (p-value = .921). There is statistical
evidence that type of coat has an effect on the body temperature of cattle (p-value = .030). There
is no statistical evidence (p-value = .066) that the strain of cattle has an effect on body
temperature.

c) Post hoc tests for Coat are appropriate because we rejected the null hypothesis that there is no
effect due to type of coat. Post hoc tests for Strain are not appropriate because we failed to reject
the null hypothesis = “there is no effect due to strain of cattle”.

2
Remark: We failed to reject the interaction hypotheses so it is appropriate to consider post
hoc tests for any main effect whose p-value ≤ α.

d) The effect on body temperature due to coat 1 differs from the effect due to coat 3. There is no
statistical evidence that the effect due to coat 1 differs from the effect due to coat 2 nor is there
statistical evidence that the effect due to coat 2 differs from the effect due to coat 3.

5) a) H 0 : σ A = σ B = σ C vs HA: at least 2 σi differ.

b) No, it is not appropriate to use an ANOVA to analyze this data set. This is not a balanced
design (there are 6 subjects receiving treatment A but 10 receive treatment B). So, we must
compare Levene’s p-value to .10. Because the p-value = .037 < .10, we conclude the
different treatment group’s standard deviations are not similar enough to use an ANOVA to
analyze this data set.s

Remark: If we were to use an ANOVA to analyze this data set, our effective type one error
rate would be greater than the stated α level.

6)
• Subjects: typists
• Factor (independent variable): background music
• Block: secretary (each subject receives each treatment so each secretary is her own
Block)
• Dependent variable: score on typing test

This is a randomized block design because every subject’s typing speed and accuracy was tested under
every treatment. When each subject’s response is measured under each treatment condition, a blocking
design must be used.

Model: X ij = µ + α i + β j + ε ij
α i = the effect due to type of background music
β j = the effect due to secretary j.

3
Hypotheses: There are 2 factors: Factor A with I levels and Factor B with J levels.
There are K ij = K > 1 response values at each ij level combination of Factor’s A and B.

There are hypotheses for each parameter in the model, except µ and εijk. For each
parameter, the null hypothesis is “no effect”

H oAB : γ ij = 0 for all ij versus H aAB : γ ij ≠ 0 for some ij


• These hypotheses test for an interaction between the 2 factors. The null
hypothesis states there is no interaction.

H oA : α1 = α 2 = ⋯ = α I = 0 versus H aA : at least one αi ≠ 0


• The null hypothesis states that the levels of factor A all have the same effect on
the response. In other words, factor A does not affect the response.

H oB : β1 = β 2 = ⋯ = β J = 0 versus H aB : at least one β j ≠ 0

Source of Sums of
variability Squares DF Mean Squares F
SSA MSA
MSA = fA =
Factor A SSA I −1 I −1 MSE
SSB MSB
MSB = fB =
Factor B SSB J −1 J −1 MSE
SSAB MSAB
Factor A* Factor B MSAB = f AB =
(Interaction term) SSAB ( I − 1)( J − 1) ( I − 1)( J − 1) MSE
SSE
MSE =
Error SSE IJ ( K − 1) IJ ( K − 1)
Total SST IJK − 1

NOTE: For all ANOVA tables, the degrees of freedom for all terms above the total must
sum to the total degrees of freedom.
• In the table above, df(factor A) + df(factor B) + df(interaction) + df(error) =df(total)

Reaching a decision based on the f statistics

• First decide whether or not to reject H oAB : γ ij = 0

o We reject H oAB if f AB ≥ Fα ,( I −1)( J −1), IJ ( K −1)

• Only if we fail to reject H oAB do we then check both factor A and factor B
hypotheses to see if either affects the dependent variable.

o We reject HoA if f A ≥ Fα , I −1, IJ ( K −1)

o We reject HoB if f B ≥ Fα , J −1, IJ ( K −1)

Вам также может понравиться