13.07.2015 Views

Consciousness-Based Education - Maharishi University of ...

Consciousness-Based Education - Maharishi University of ...

Consciousness-Based Education - Maharishi University of ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

U.S.-Soviet Relations and the <strong>Maharishi</strong> Effect1981). The ADF test was based on the ADF regression without trendor intercept. The latter two parameters were found to be not significantusing tabled critical values in Dickey and Fuller (1981) (n = 100).The insignificance <strong>of</strong> the trend and intercept implies that maximumpower for the ADF test will be obtained using the regression excludingthese parameters. To correct for serial correlation, five lagged firstdifferencedterms for the dependent variable were included in the ADFregression for the USSR tand two lagged terms for the ADF regressionfor USt. Lagrange Multiplier tests for residual autocorrelation at lags1–2 and lags 1–12 were not significant for both ADF regressions.10Unlike the ADF test, which rests on the assumption <strong>of</strong> independentand identically distributed residual errors in the ADF regression, thePhillips-Perron generalization <strong>of</strong> the Dickey-Fuller procedure remainsappropriate even under the weaker assumption that the errors areweakly dependent and heterogeneously distributed (Phillips and Perron,1988).11In the Phillips-Perron procedure, the t-statistic <strong>of</strong> the estimatedregression coefficients are corrected for possible serial correlation inthe disturbance u tusing the Newey-West method for adjusting thestandard errors (Newey and West, 1987). Using a formula suggestedby Newey and West, the truncation lag q was set equal to 3 in theBartlett kernel used to approximate the dynamics <strong>of</strong> the residuals forthe purpose <strong>of</strong> correcting for serial correlation. This parameter definesthe number <strong>of</strong> autocorrelations using in the adjustment procedure andwas calculated as q = 4(n/100) 2/9 .12These tabled critical values are the same for both the Phillips-Perronand the augmented Dickey-Fuller tests.13To maximize the power <strong>of</strong> the test to reject the null hypothesis <strong>of</strong>nonstationarity, a sequential testing procedure (Doldado, Jenkinson,and Sosvilla-Rivero, 1990) may be used if the null hypothesis <strong>of</strong> a unitroot is not rejected for the initial PP (or ADF) regression that includesboth trend and intercept terms.14Maximum likelihood estimates were calculated using SCA StatisticalSystem s<strong>of</strong>tware for the PC.15Since the t-distribution approaches the standard normal as samplesize increases, use <strong>of</strong> p-values based on the t-distribution for samplesize n = 93 would not materially change any <strong>of</strong> the statistical inferences487

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!