1
$\begingroup$

At a measurement station in Munich for 14 days the sulfur dioxide concentration was measured. Analyze the influence of the mean temperature per day in grad Celsius ($X_1$) on the $SO_2$-concentration $(Y)$. Is there a special effect on weekends? The variable $X_2$ is $1$ if it was measured on a saturday or sunday otherwise it is $0$. The following data were measured: $$ \begin{array}{c|c|c|c|c|c|c|c} \hline y & -3.15 & -2.83 & -3.02 & -3.08 & -3.54 & -2.98 & -2.78\\ x_1 & 16.47 & 16.02 & 16.81 & 22.87 & 21.68 & 21.23 & 20.55\\ x_2 & 0 & 0 & 0 & 1 & 1 & 0 & 0\\ \hline \end{array} $$ \begin{array}{c|c|c|c|c|c|c|c} \hline y & -3.35 & -2.76 & -1.90 & -2.12 & -2.45 & -1.97 & -2.23\\ x_1 & 18.32 & 15.96 & 15.36 & 12.47 & 12.46 & 11.77 & 11.72\\ x_2 & 0 & 0 & 0 & 1 & 1 & 0 & 0 \end{array} (a) Estimate the regression coefficients of the multiple linear model and comment your results. (b) As $R^2$ one gets $R^2=0.5805$. Does the regressors have any role by explaining the $SO_2$-concentration? Make a global F-test of level $\alpha=0.01$. (c) The estimated variances are $\hat{\sigma}_1=0.0267$ and $\hat{\sigma}_2=0.2166$. Test the hypothesis $\theta_i=0$ for $i=2,3$ to the level $\alpha=0.05$ by confidence intervalls. Remove the covariable that does not have an influence and then make a simple linear regression. Compare the multiple and the simple regression graphically.

Hello and good day!

I managed most things of this task already.

First of all I examined by ANOVA if there is a weekend effect. I chose $\alpha=0.05$. For the $F$-statstic I got $F=0.1$ and a p-value of $0.76$. So the null hypothesis that $\theta_1=\theta_2$ can not be fefused and so there is no weekend effect.

(a) The estimated coefficients are $$ \hat{\theta}_1=-1.0019,~~~\hat{\theta}_2=-0.1032,~~~\hat{\theta}_3=-0.0025, $$ where the first coefficient is the intercept, the second coefficient belongs to $X_1$ and the thirds one to $X_2$.

Now I do not know what I should comment here? What could be meant? What is to be commented?

(b)

The global $F$-test says that the hypothesis $H_0: \theta_2=\theta_3=0$ can be refused. So the regressors play a role for explaining the $SO_2$-concentration.

(c)

For $\theta_2$ the confidence interval is $$ C_2^{0.05}=(-0.156863,-0.049537), $$ so because $0\notin C_{2}^{0.05}$, the hypothesis $H_0^2: \theta_2=0$ can be refused.

For $\theta_3$ the confidence intervall is given by $$ C_3^{0.05}=(-0.437834,0.432834), $$ so because of $0\in C_3^{0.05}$, the hypothesis $H_0^3: \theta_3=0$ cannot be refused, so the covariable $X_3$ does not have an influence and can be removed from the model.

I now made the simple linear regression, again using R, getting now $$ \hat{\theta}_1=-1.0020,~~~\hat{\theta}_2=-0.1033. $$

What I see is, that the coefficients are nearly the same compared to those of the multiple regression.

But I do not know how to compare the simple linear regression and the multiple linear regression graphically. Of course, I can plot the simple linear regression:

simple linear regression

but I do not know how to plot the multiple regression in order to compare it then with the simple linear regression. Can you help me with that?


So to sum it up:

Concerning (a) I do not know what is to be commented and concerning (c) I do not know how I can compare the multiple and the simple regression graphically.

Would be great if you could help me with that!

See you

Miroslav

$\endgroup$
  • $\begingroup$ I'm stuck just trying to understand the data. What do you suppose those negative "concentrations" mean? $\endgroup$ – whuber Jun 10 '14 at 18:37
  • 2
    $\begingroup$ The data of $Y$ were logarithmized. $\endgroup$ – mathfemi Jun 10 '14 at 19:21
0
$\begingroup$

Miroslav,

I'm an engineer, not a pure-math guy. I aspire to pure math.

When I hear the question, my concrete physical reality self kicks in and asks it slightly differently. I hear "does the answer make sense" and "what do the numbers mean". What does it imply if there is a weekend effect?

When I want to compare - I would look at error histograms. I would compute error for the simple regression, and look at the distribution plot. (link) I would then compute error for the multiple regression and look at the distribution plot. Using these I could look at typical error, maximum error between the fits.

I'm wondering if the negative value is related to the idea of pH. If there is a logarithm of a concentration between 0 and 100% then a negative prefix would be required for anything but pure sulfur dioxide. For all $x<1$ the value of $log_{10}\left(x\right)$ is negatively valued. When I check this (link) then I find that 800 micrograms per cubic meter is an alert-level of sulfur dioxide. If I compute $ 10^{-y}$ I can get on the order of ~800 units for some of the values. The measurement is one per this, so around 1 part per 800. I think that they compute how many micrograms per cubic meter, then take the $log_{10}$ of that.

Best of luck.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.