Pages

Thursday, December 07, 2017

Optimistic Findings about California Air Pollution Trends During the Fire Season

The recent Los Angeles fires have been quite scary.  When I'm scared, I start to run new regressions.  I take daily PM2.5 air pollution data from the EPA and keep the subset of observations for the following states; California, Arizona and Nevada.  I use data from the years 2000 to 2017.   I merge in data on daily wind speed and average temperature.  I create two key variables;

Fireseason = 1 if the day is in September, October, November, December

trend  =  the monthly time trend (so there are 17 years and 12 months per year so this variable takes on the values 1 to 12*17)

inter  =  Fireseason*trend   , this interaction term tests if the monthly time trend differs during the fire season.

So, for 152 monitoring stations in Arizona, California and Nevada; I am running a regression where the Y variable is the log(PM2.5) on a given day.  I include monitoring station fixed effects and I regress this on a monthly time trend (t),  the fireseason dummy (see above) the time trend during the fireseason months and the climate variables (log wind and log average temperature).



. areg lmean t inter  fireseason lwind lavgtemp , absorb(idno) cluster(idno)

Linear regression, absorbing indicators         Number of obs     =    405,388
                                                F(   5,    151)   =      47.76
                                                Prob > F          =     0.0000
                                                R-squared         =     0.2775
                                                Adj R-squared     =     0.2772
                                                Root MSE          =     0.6412

                                 (Std. Err. adjusted for 152 clusters in idno)
------------------------------------------------------------------------------
             |               Robust
       lmean |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
           t |  -.0018944   .0002363    -8.02   0.000    -.0023612   -.0014276
       inter |   -.001042   .0002175    -4.79   0.000    -.0014717   -.0006123
  fireseason |   .2643525   .0310733     8.51   0.000     .2029578    .3257471
       lwind |  -.1425646   .0743011    -1.92   0.057    -.2893687    .0042395
    lavgtemp |  -.0306948    .072866    -0.42   0.674    -.1746633    .1132738
       _cons |   3.133668   .5095623     6.15   0.000     2.126875     4.14046
-------------+----------------------------------------------------------------
        idno |   absorbed                                     (152 categories)


What have we learned?    All else equal, West coast PM2.5 is 26.4% higher during the fire months.  But, for those who are looking for optimism --- note the time trends.  The monthly time trend = -.0018 , so for every 10 months that pass; PM2.5 declines by 1.8% in the non-fire season months. Note that inter has a negative coefficient of -.001.  During the fire season months;  the downward trend in PM2.5 is even more pronounced.    During the fireseason, PM2.5 is trending down at about 3% per year.

These results are roughly the same when I weight by county population size (so I'm placing more importance on LA county). Note the time trend is even more negative in this case.




. areg lmean t inter  fireseason lwind lavgtemp  [w=pop2000], absorb(idno) clus
> ter(idno)
(analytic weights assumed)
(analytic weights assumed)
(sum of wgt is   6.6541e+11)

Linear regression, absorbing indicators         Number of obs     =    405,388
                                                F(   5,    151)   =      64.11
                                                Prob > F          =     0.0000
                                                R-squared         =     0.3227
                                                Adj R-squared     =     0.3224
                                                Root MSE          =     0.5881

                                 (Std. Err. adjusted for 152 clusters in idno)
------------------------------------------------------------------------------
             |               Robust
       lmean |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
           t |  -.0024729   .0002347   -10.54   0.000    -.0029366   -.0020092
       inter |  -.0010034   .0002073    -4.84   0.000     -.001413   -.0005938
  fireseason |   .2544869   .0254936     9.98   0.000     .2041168     .304857
       lwind |   -.002755   .0483044    -0.06   0.955    -.0981948    .0926848
    lavgtemp |   .0321614   .1147983     0.28   0.780    -.1946571    .2589798
       _cons |   2.422674   .5416897     4.47   0.000     1.352404    3.492944
-------------+----------------------------------------------------------------
        idno |   absorbed                                     (152 categories)


Fires are certainly horrible but the drought prone region is suffering less PM2.5 pollution over time.  These regressions cannot tell you why this is the case.

Here are the results just for California.


. areg lmean t inter  fireseason lwind lavgtemp if state==6, absorb(idno) clust
> er(idno)

Linear regression, absorbing indicators         Number of obs     =    318,986
                                                F(   5,    118)   =      53.93
                                                Prob > F          =     0.0000
                                                R-squared         =     0.2530
                                                Adj R-squared     =     0.2527
                                                Root MSE          =     0.6482

                                 (Std. Err. adjusted for 119 clusters in idno)
------------------------------------------------------------------------------
             |               Robust
       lmean |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
           t |  -.0017874   .0002511    -7.12   0.000    -.0022846   -.0012902
       inter |  -.0012424   .0002473    -5.02   0.000    -.0017321   -.0007527
  fireseason |   .2895557   .0306528     9.45   0.000     .2288548    .3502565
       lwind |  -.1454173   .0835908    -1.74   0.085    -.3109498    .0201152
    lavgtemp |     .03904   .0957566     0.41   0.684    -.1505842    .2286641
       _cons |   2.894797   .6436804     4.50   0.000     1.620135    4.169459
-------------+----------------------------------------------------------------
        idno |   absorbed                                     (119 categories)