Research Research Paper Samples

The goal of this research is to construct a multiple linear regression equation between

YEARSJOB (dependent variable) and JOBLOSE JOBSAFE JOBKEEP SATFIN FINALTER UNEMP (independent variables).
I begin with a descriptive statistics of this data:
. summarize yearsjob joblose satfin finalter unemp
Variable | Obs Mean Std. Dev. Min Max
yearsjob | 1156 7.956315 8.906898 .25 50
joblose | 801 3.292135 .8927769 1 4
satfin | 2038 2.080471 .7371224 1 3
finalter | 2033 2.132317 .7828774 1 3
unemp | 1374 1.621543 .485179 1 2
The data can be visualized by frequency histograms. We want to know if the data distribution significantly different from a normal distribution since it is one of the assumption of the regression analysis.
. histogram yearsjob, frequency
. histogram joblose, frequency
. histogram satfin, frequency
. histogram finalter, frequency
. histogram unemp, frequency
It seems that the data of yearsjob is very positively skewed, the data of other variables are not normal at all. This fact may bias our results. However, we proceed with a regression analysis:
. regress yearsjob joblose satfin finalter unemp
Source | SS df MS Number of obs = 397
-------------+------------------------------ F( 4, 392) = 30.40
Model | 7122.49811 4 1780.62453 Prob > F = 0.0000
Residual | 22957.3728 392 58.5647265 R-squared = 0.2368
-------------+------------------------------ Adj R-squared = 0.2290
Total | 30079.8709 396 75.95927 Root MSE = 7.6528
yearsjob | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
joblose | .5361403 .4622294 1.16 0.247 -.3726185 1.444899
satfin | -.4877709 .5573634 -0.88 0.382 -1.583566 .6080246
finalter | 1.734354 .4701975 3.69 0.000 .8099296 2.658778
unemp | 7.76058 .8228909 9.43 0.000 6.142748 9.378411
_cons | -8.930296 2.574633 -3.47 0.001 -13.99211 -3.868479
According to the ANOVA result (F=30.4, p<0.001), the overall model is significant and can be used in making forecasts. However, some factors are not significant. The p-values of joblose and satfin are higher than 0.05, that’s why I consider them insignificant in relation to affect the dependent variable. That’s why they should be removed from the model.
The overall model goodness is low, because the coefficient of determination R-squared is only 0.2368. This means that only 23.68% of variance of yearsjob is explained by this model.

Exclude the variables mentioned above and make a regression again:

. regress yearsjob finalter unemp
Source | SS df MS Number of obs = 759
-------------+------------------------------ F( 2, 756) = 94.17
Model | 11645.8971 2 5822.94854 Prob > F = 0.0000
Residual | 46744.3639 756 61.8311692 R-squared = 0.1994
-------------+------------------------------ Adj R-squared = 0.1973
Total | 58390.261 758 77.0320066 Root MSE = 7.8633
yearsjob | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
finalter | .960918 .3513612 2.73 0.006 .2711585 1.650678
unemp | 7.842864 .5877776 13.34 0.000 6.688994 8.996734
_cons | -6.628035 1.201942 -5.51 0.000 -8.987576 -4.268494
Now, the overall model is still significant and each factor included in the model is also significant. R-squared now is lower (19.94%).
Generally, this equation is better than the one mentioned above, but it is still weak, since R-squared is far from 1. There must be some factors which have a significant impact on yearsjob and they are not included in this model.