In the above example we had no reason to suspect that either population watched more TV than the other, as reflected in the alternative hypothesis, which is simply a statement of difference:
µ1 ≠ µ2
Given this formulation of the alternative hypothesis, a two-tail test is used. However, we might argue that because of the warmer weather Australian children might be less inclined to watch as much TV as children in Britain. Here we hot only suspect a difference, but also a direction of difference. In this situation the alternative hypothesis will be:
µ1 ˃ µ2
This would involve a one-tail test. The SPSS output provides the two-tail significance as the default setting. To convert this into a one-tail significance we simply divide the two-tail probability in half. The area under the curve beyond a t-score of ± 2.25 is 0.031 of the total area under the t-distribution. If two-tail take up this amount of area, then one-tail will take up half this amount:
One-tail probability = (two tail probabillity)/2
= 0.031/2
= 0.0155
Example
A study is conducted to investigate whether foreign companies pay attention to local health and safety codes compared with domestic companies. A survey of 50 foreign-owned and 50 domestic companies of similar industries are selected. Inspectors record the number of breaches of health and safety regulations they observe when inspecting these establishments. On average, the 50 foreign firms were found to make 4.2 breaches per firm:
Foreign: N1 = 50
S1 = 1.3
x̄ = 4.2
The domestic firms were found to make 3.5 breaches per firm
Domestic: N2 = 50
s2 = 1.2
x̄ = 3.5
We will use an alpha level of 0.05, which on a two-tail test, and with 98 degrees of freedom, gives us a critical score of:
tcritical = ±2.0
To calculate the sample t-score we need to firstly calculate the standard error (using the equal variance estimate):
σ x̄ - x̄ =
=
= 0.25
The sample t-score will be:
tsample = (x ̄1-x̄2 )/(σ x̄ - x̄ )
= (4.2-3.5)/0.25
= 2.8
We can see that sample score is further from zero than the critical score. In other words, it falls in the region of rejection (Figure 13.7).
Figure 13.7 Distribution of t, df = 98
We reject the null hypothesis of no difference. The results suggest that foreign firms are more inclined to breach local health and safety codes.