You've collected your data, you've got your model, you've run your regression and you've got your results. Now what do you do with your results?
In this article we consider the Okun's Law model and results from the article "How to Do a Painless Econometrics Project". One sample t-tests will be introduced and used in order to see if the theory matches the data.
The theory behind Okun's Law was described in the article: "Instant Econometrics Project 1 - Okun's Law":
Okun's law is an empirical relationship between the change in the unemployment rate and the percentage growth in real output, as measured by GNP. Arthur Okun estimated the following relationship between the two:
Y_{t} = - 0.4 (X_{t} - 2.5 )
This can also be expressed as a more traditional linear regression as:
Y_{t} = 1 - 0.4 X_{t}
Where:
Y_{t} is the change in the unemployment rate in percentage points.
X_{t} is the percentage growth rate in real output, as measured by real GNP.
So our theory is that the values of our parameters are B_{1} = 1 for the slope parameter and B_{2} = -0.4 for the intercept parameter.
We used American data to see how well the data matched the theory. From "How to Do a Painless Econometrics Project" we saw that we needed to estimate the model:
Y_{t} = b_{1} + b_{2} X_{t}
Y_{t} X_{t} b_{1} b_{2} B_{1} B_{2}Using Microsoft Excel, we calculated the parameters b_{1} and b_{2}. Now we need to see if those parameters match our theory, which was that B_{1} = 1 and B_{2} = -0.4. Before we can do that, we need to jot down some figures that Excel gave us. If you look at the results screenshot you'll notice that the values are missing. That was intentional, as I want you to calculate the values on your own. For the purposes of this article, I will make up some values and show you in what cells you can find the real values. Before we begin our hypothesis testing, we need to jot down the following values:
Observations
- Number of Observations (Cell B8) Obs = 219
Intercept
- Coefficient (Cell B17) b_{1} = 0.47 (appears on chart as "AAA")
Standard Error (Cell C17) se_{1} = 0.23 (appears on chart as "CCC")
t Stat (Cell D17) t_{1} = 2.0435 (appears on chart as "x")
P-value (Cell E17) p_{1} = 0.0422 (appears on chart as "x")
X Variable
- Coefficient (Cell B18) b_{2} = - 0.31 (appears on chart as "BBB")
Standard Error (Cell C18) se_{2} = 0.03 (appears on chart as "DDD")
t Stat (Cell D18) t_{2} = 10.333 (appears on chart as "x")
P-value (Cell E18) p_{2} = 0.0001 (appears on chart as "x")
In the next section we'll look at hypothesis testing and we'll see if our data matches our theory.
Be Sure to Continue to Page 2 of "Hypothesis Testing Using One-Sample t-Tests".
First we’ll consider our hypothesis that the intercept variable equals one. The idea behind this is explained quite well in Gujarati’s Essentials of Econometrics. On page 105 Gujarati describes hypothesis testing:
- “[S]uppose we hypothesize that the true B_{1} takes a particular numerical value, e.g., B_{1} = 1. Our task now is to “test” this hypothesis.”“In the language of hypothesis testing a hypothesis such as B_{1} = 1 is called the null hypothesis and is generally denoted by the symbol H_{0}. Thus H_{0}: B_{1} = 1. The null hypothesis is usually tested against an alternative hypothesis, denoted by the symbol H_{1}. The alternative hypothesis can take one of three forms:
H_{1}: B_{1} > 1, which is called a one-sided alternative hypothesis, or
H_{1}: B_{1} < 1, also a one-sided alternative hypothesis, or
H_{1}: B_{1} not equal 1, which is called a two-sided alternative hypothesis. That is the true value is either greater or less than 1.”
In the above I’ve substituted in our hypothesis for Gujarati’s to make it easier to follow. In our case we want a two-sided alternative hypothesis, as we’re interested in knowing if B_{1} is equal to 1 or not equal to 1.
The first thing we need to do to test our hypothesis is to calculate at t-Test statistic. The theory behind the statistic is beyond the scope of this article. Essentially what we are doing is calculating a statistic which can be tested against a t distribution to determine how probable it is that the true value of the coefficient is equal to some hypothesized value. When our hypothesis is B_{1} = 1 we denote our t-Statistic as t_{1}(B_{1}=1) and it can be calculated by the formula:
t_{1}(B_{1}=1) = (b_{1} - B_{1} / se_{1})
Let’s try this for our intercept data. Recall we had the following data:
Intercept
- b_{1} = 0.47
se_{1} = 0.23
Our t-Statistic for the hypothesis that B_{1} = 1 is simply:
t_{1}(B_{1}=1) = (0.47 – 1) / 0.23 = 2.0435
So t_{1}(B_{1}=1) is 2.0435. We can also calculate our t-test for the hypothesis that the slope variable is equal to -0.4:
X Variable
- b_{2} = -0.31
se_{2} = 0.03
Our t-Statistic for the hypothesis that B_{2} = -0.4 is simply:
t_{2}(B_{2}= -0.4) = ((-0.31) – (-0.4)) / 0.23 = 3.0000
So t_{2}(B_{2}= -0.4) is 3.0000. Next we have to convert these into p-values. The p-value "may be defined as the lowest significance level at which a null hypothesis can be rejected...As a rule, the smaller the p value, the stronger is the evidence against the null hypothesis." (Gujarati, 113) As a standard rule of thumb, if the p-value is lower than 0.05, we reject the null hypothesis and accept the alternative hypothesis. This means that if the p-value associated with the test t_{1}(B_{1}=1) is less than 0.05 we reject the hypothesis that B_{1}=1 and accept the hypothesis that B_{1} not equal to 1. If the associated p-value is equal to or greater than 0.05, we do just the opposite, that is we accept the null hypothesis that B_{1}=1.
Calculating the p-value
Unfortunately, you cannot calculate the p-value. To obtain a p-value, you generally have to look it up in a chart. Most standard statistics and econometrics books contain a p-value chart in the back of the book. Fortunately with the advent of the internet, there’s a much simpler way of obtaining p-values. The site Graphpad Quickcalcs: One sample t test allows you to quickly and easily obtain p-values. Using this site, here’s how you obtain a p-value for each test.
Steps Needed to Estimate a p-value for B_{1}=1
- Click on the radio box containing “Enter mean, SEM and N.” Mean is the parameter value we estimated, SEM is the standard error, and N is the number of observations.
- Enter 0.47 in the box labelled “Mean:”.
- Enter 0.23 in the box labelled “SEM:”
- Enter 219 in the box labelled “N:”, as this is the number of observations we had.
- Under " 3. Specify the hypothetical mean value" click on the radio button beside the blank box. In that box enter 1, as that is our hypothesis.
- Click “Calculate Now”
You should get an output page. On the top of the output page you should see the following information:
- P value and statistical significance:
The two-tailed P value equals 0.0221
By conventional criteria, this difference is considered to be statistically significant.
So our p-value is 0.0221 which is less than 0.05. In this case we reject our null hypothesis and accept our alternative hypothesis. In our words, for this parameter, our theory did not match the data.
Be Sure to Continue to Page 3 of "Hypothesis Testing Using One-Sample t-Tests".
Again using site Graphpad Quickcalcs: One sample t test we can quickly obtain the p-value for our second hypothesis test:
Steps Needed to Estimate a p-value for B_{2}= -0.4
- Click on the radio box containing Enter mean, SEM and N. Mean is the parameter value we estimated, SEM is the standard error, and N is the number of observations.
- Enter -0.31 in the box labelled Mean:.
- Enter 0.03 in the box labelled SEM:
- Enter 219 in the box labelled N:, as this is the number of observations we had.
- Under 3. Specify the hypothetical mean value click on the radio button beside the blank box. In that box enter -0.4, as that is our hypothesis.
- Click Calculate Now
- P value and statistical significance: The two-tailed P value equals 0.0030
By conventional criteria, this difference is considered to be statistically significant.
We used U.S. data to estimate the Okun's Law model. Using that data we found that both the intercept and slope parameters are statistically significantly different than those in Okun's Law. Therefore we can conclude that in the United States Okun's Law does not hold.
Now you've seen how to calculate and use one-sample t-tests, you will be able to interpret the numbers you've calculated in your regression.
If you'd like to ask a question about econometrics, hypothesis testing, or any other topic or comment on this story, please use the feedback form. If you're interested in winning cash for your economics term paper or article, be sure to check out "The 2004 Moffatt Prize in Economic Writing"