Reply #53: More careful analysis

[xocolatl]

OK, so it turns out that your claim appears to have merit, given a more careful examination of the raw data. The P-value turns out to be 0.002.

Using a crossed-effects model (with random month and news-outlet effects), and a single fixed effect "Gp" (=1 for FOX, AP, and CNN, =0 otherwise):

Solution for Fixed Effects

Standard
Effect Estimate Error DF t Value Pr > |t|

Intercept 1.4360 1.8789 6 0.76 0.4737
Gp -4.8344 1.4882 52 -3.25 0.0020


Covariance Parameter Estimates

Cov Parm Estimate
News 2.1626
Mon 20.8829
Residual 16.7650


More interesting are the empirical BLPs for the random effects:


Std Err
Effect News Mon Estimate Pred DF t Value Pr > |t|

News ABC 1.0692 1.1872 52 0.90 0.3719
News AP -0.6102 1.2565 52 -0.49 0.6293
News ARG 1.0133 1.1223 52 0.90 0.3708
News CBS 0.8099 1.1223 52 0.72 0.4737
News CNN 1.2737 1.2298 52 1.04 0.3051
News FOX -0.6635 1.2416 52 -0.53 0.5954
News IBD -0.8934 1.1523 52 -0.78 0.4416
News LA 0.6055 1.3281 52 0.46 0.6503
News NBC -1.5324 1.1886 52 -1.29 0.2030
News NWK 0.2028 1.1541 52 0.18 0.8612
News PEW -0.5912 1.1532 52 -0.51 0.6104
News TIME -0.5247 1.2299 52 -0.43 0.6714
News ZOGBY -0.1591 1.1871 52 -0.13 0.8939
Mon Apr 0.8177 2.0586 52 0.40 0.6929
Mon Feb 0.8211 2.1102 52 0.39 0.6988
Mon Jan -9.6705 2.1118 52 -4.58 <.0001
Mon July 3.4853 2.1116 52 1.65 0.1049
Mon June 0.4052 2.0846 52 0.19 0.8466
Mon Mar 2.3544 2.0561 52 1.15 0.2574
Mon May 1.7869 2.0356 52 0.88 0.3841


I can send SAS code and raw data to your DU email if you wish.