Re: [R] Exact p-values in lm() - rounding problem
On 12.02.2013 13:37, Torvon wrote: I need to report exact p-values in my dissertation. Looking at my lm() results of many regressions with huge datasets I have the feeling that p-values are rounded to the smallest value of 2e-16, because this p-value is very common. Is that true or just chance? If it is true, how do I obtain the true unrounded p-values for these regressors? m1 - lm(y ~ x1+x2+x3+4+x5, data=D) coef(summary(m1))[,4] Anyway, you should not believe that smaller values are still accurate. Always worry about the numerics when looking at tiny differences. Best, Uwe Ligges Thank you Torvon [[alternative HTML version deleted]] __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
Re: [R] Exact p-values in lm() - rounding problem
Thank you, Uwe. summary(m1) gives me p-value estimates of: (Intercept) 2e-16 x1 6.9e-15 x2 1.9e-07 x3 2.7e-09 While coef(summary(m1))[,4] gives me: (Intercept) 3.0e-23 x1 5.7e-13 x2 2.6e-07 x3 1.7e-17 While the first one confirms my suspicion (-23 instead of -16), the latter one vary drastically (especially x3 from -09 to -17). Why is that? Thank you! T. [[alternative HTML version deleted]] __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
Re: [R] Exact p-values in lm() - rounding problem
Torvon torvon at gmail.com writes: Thank you, Uwe. summary(m1) gives me p-value estimates of: (Intercept) 2e-16 x1 6.9e-15 x2 1.9e-07 x3 2.7e-09 While coef(summary(m1))[,4] gives me: (Intercept) 3.0e-23 x1 5.7e-13 x2 2.6e-07 x3 1.7e-17 While the first one confirms my suspicion (-23 instead of -16), the latter one vary drastically (especially x3 from -09 to -17). Why is that? This looks fishy. Could we please have a reproducible example? For example, give us the results of dput(summary(m1)) ... When I try this with example(lm) it works fine: Coefficients: Estimate Std. Error t value Pr(|t|) (Intercept) 5.0320 0.2202 22.850 9.55e-15 *** groupTrt -0.3710 0.3114 -1.1910.249 coef(summary(lm.D9))[,4] (Intercept) groupTrt 9.547128e-15 2.490232e-01 __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
Re: [R] Exact p-values in lm() - rounding problem
On 12.02.2013 14:44, Torvon wrote: Thank you, Uwe. summary(m1) gives me p-value estimates of: (Intercept) 2e-16 x1 6.9e-15 x2 1.9e-07 x3 2.7e-09 While coef(summary(m1))[,4] gives me: (Intercept) 3.0e-23 x1 5.7e-13 x2 2.6e-07 x3 1.7e-17 While the first one confirms my suspicion (-23 instead of -16), the latter one vary drastically (especially x3 from -09 to -17). Why is that? Can you show the complete code and output? Uwe Ligges Thank you! T. __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
Re: [R] Exact p-values in lm() - rounding problem
The code is quite long because I am running a WLS regression instead of an OLS regression (due to heteroscedasticity). First, I get mean structure, then get mean/SD relationship, then improve the variance structure by using weights proportional to 1/variance. I am quite sure this is not relevant, so I will only post the rest of the code. Let me know if you need that part, too. I appreciate the help Uwe! Best, T. m3 = lm(s8_1234_m~ Sex + HisDep + FamHis + ZNeuro + ZEFE + Zwh_1234_m + Zale_1234_m+t0s8, weights=W, data=D) summary(m3) Call: lm(formula = s8_1234_m ~ Sex + HisDep + FamHis + ZNeuro + ZEFE + Zwh_1234_m + Zale_1234_m + t0s8, data = D, weights = W) Residuals: Min 1Q Median 3Q Max -1.3691 -0.5453 -0.4104 0.2606 7.0111 Coefficients: Estimate Std. Error t value Pr(|t|) (Intercept) 0.209610.01681 12.472 2e-16 *** Sex -0.023210.01708 -1.359 0.17435 HisDep 0.025440.01987 1.281 0.20052 FamHis -0.021830.01798 -1.215 0.22478 ZNeuro 0.079390.01007 7.882 6.87e-15 *** ZEFE 0.022430.01056 2.124 0.03385 * Zwh_1234_m 0.042650.00814 5.240 1.88e-07 *** Zale_1234_m 0.028770.00975 2.951 0.00323 ** t0s8 0.389800.06504 5.993 2.67e-09 *** --- Signif. codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.1 1 Residual standard error: 0.9321 on 1280 degrees of freedom Multiple R-squared: 0.1282, Adjusted R-squared: 0.1228 F-statistic: 23.54 on 8 and 1280 DF, p-value: 2.2e-16 coef(summary(m1))[,4] (Intercept) Sex HisDep FamHis ZNeuro ZEFE Zwh_1234_m Zale_1234_m 3.042584e-23 2.146371e-01 2.769561e-01 9.988154e-01 5.682278e-13 5.243800e-03 2.599513e-07 3.116738e-02 t0s8 1.741608e-17 On 12 February 2013 15:07, Uwe Ligges lig...@statistik.tu-dortmund.dewrote: On 12.02.2013 14:44, Torvon wrote: Thank you, Uwe. summary(m1) gives me p-value estimates of: (Intercept) 2e-16 x1 6.9e-15 x2 1.9e-07 x3 2.7e-09 While coef(summary(m1))[,4] gives me: (Intercept) 3.0e-23 x1 5.7e-13 x2 2.6e-07 x3 1.7e-17 While the first one confirms my suspicion (-23 instead of -16), the latter one vary drastically (especially x3 from -09 to -17). Why is that? Can you show the complete code and output? Uwe Ligges Thank you! T. [[alternative HTML version deleted]] __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
Re: [R] Exact p-values in lm() - rounding problem
On 12.02.2013 15:15, Torvon wrote: The code is quite long because I am running a WLS regression instead of an OLS regression (due to heteroscedasticity). First, I get mean structure, then get mean/SD relationship, then improve the variance structure by using weights proportional to 1/variance. I am quite sure this is not relevant, so I will only post the rest of the code. Let me know if you need that part, too. I appreciate the help Uwe! Best, T. m3 = lm(s8_1234_m~ Sex + HisDep + FamHis + ZNeuro + ZEFE + Zwh_1234_m + Zale_1234_m+t0s8, weights=W, data=D) summary(m3) Call: lm(formula = s8_1234_m ~ Sex + HisDep + FamHis + ZNeuro + ZEFE + Zwh_1234_m + Zale_1234_m + t0s8, data = D, weights = W) Residuals: Min 1Q Median 3Q Max -1.3691 -0.5453 -0.4104 0.2606 7.0111 Coefficients: Estimate Std. Error t value Pr(|t|) (Intercept) 0.209610.01681 12.472 2e-16 *** Sex -0.023210.01708 -1.359 0.17435 HisDep 0.025440.01987 1.281 0.20052 FamHis -0.021830.01798 -1.215 0.22478 ZNeuro 0.079390.01007 7.882 6.87e-15 *** ZEFE 0.022430.01056 2.124 0.03385 * Zwh_1234_m 0.042650.00814 5.240 1.88e-07 *** Zale_1234_m 0.028770.00975 2.951 0.00323 ** t0s8 0.389800.06504 5.993 2.67e-09 *** --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 0.9321 on 1280 degrees of freedom Multiple R-squared: 0.1282, Adjusted R-squared: 0.1228 F-statistic: 23.54 on 8 and 1280 DF, p-value: 2.2e-16 coef(summary(m1))[,4] (Intercept) Sex HisDep FamHis ZNeuro ZEFE Zwh_1234_m Zale_1234_m 3.042584e-23 2.146371e-01 2.769561e-01 9.988154e-01 5.682278e-13 5.243800e-03 2.599513e-07 3.116738e-02 t0s8 1.741608e-17 So you are comparing results from m3 with those from m1 Uwe Ligges On 12 February 2013 15:07, Uwe Ligges lig...@statistik.tu-dortmund.dewrote: On 12.02.2013 14:44, Torvon wrote: Thank you, Uwe. summary(m1) gives me p-value estimates of: (Intercept) 2e-16 x1 6.9e-15 x2 1.9e-07 x3 2.7e-09 While coef(summary(m1))[,4] gives me: (Intercept) 3.0e-23 x1 5.7e-13 x2 2.6e-07 x3 1.7e-17 While the first one confirms my suspicion (-23 instead of -16), the latter one vary drastically (especially x3 from -09 to -17). Why is that? Can you show the complete code and output? Uwe Ligges Thank you! T. [[alternative HTML version deleted]] __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
Re: [R] Exact p-values in lm() - rounding problem
I'm sorry, no clue how I did not see that. Thank you! On 12 February 2013 15:21, Uwe Ligges lig...@statistik.tu-dortmund.dewrote: On 12.02.2013 15:15, Torvon wrote: The code is quite long because I am running a WLS regression instead of an OLS regression (due to heteroscedasticity). First, I get mean structure, then get mean/SD relationship, then improve the variance structure by using weights proportional to 1/variance. I am quite sure this is not relevant, so I will only post the rest of the code. Let me know if you need that part, too. I appreciate the help Uwe! Best, T. m3 = lm(s8_1234_m~ Sex + HisDep + FamHis + ZNeuro + ZEFE + Zwh_1234_m + Zale_1234_m+t0s8, weights=W, data=D) summary(m3) Call: lm(formula = s8_1234_m ~ Sex + HisDep + FamHis + ZNeuro + ZEFE + Zwh_1234_m + Zale_1234_m + t0s8, data = D, weights = W) Residuals: Min 1Q Median 3Q Max -1.3691 -0.5453 -0.4104 0.2606 7.0111 Coefficients: Estimate Std. Error t value Pr(|t|) (Intercept) 0.209610.01681 12.472 2e-16 *** Sex -0.023210.01708 -1.359 0.17435 HisDep 0.025440.01987 1.281 0.20052 FamHis -0.021830.01798 -1.215 0.22478 ZNeuro 0.079390.01007 7.882 6.87e-15 *** ZEFE 0.022430.01056 2.124 0.03385 * Zwh_1234_m 0.042650.00814 5.240 1.88e-07 *** Zale_1234_m 0.028770.00975 2.951 0.00323 ** t0s8 0.389800.06504 5.993 2.67e-09 *** --- Signif. codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.1 1 Residual standard error: 0.9321 on 1280 degrees of freedom Multiple R-squared: 0.1282, Adjusted R-squared: 0.1228 F-statistic: 23.54 on 8 and 1280 DF, p-value: 2.2e-16 coef(summary(m1))[,4] (Intercept) Sex HisDep FamHis ZNeuro ZEFE Zwh_1234_m Zale_1234_m 3.042584e-23 2.146371e-01 2.769561e-01 9.988154e-01 5.682278e-13 5.243800e-03 2.599513e-07 3.116738e-02 t0s8 1.741608e-17 So you are comparing results from m3 with those from m1 Uwe Ligges On 12 February 2013 15:07, Uwe Ligges lig...@statistik.tu-dortmund.**delig...@statistik.tu-dortmund.de wrote: On 12.02.2013 14:44, Torvon wrote: Thank you, Uwe. summary(m1) gives me p-value estimates of: (Intercept) 2e-16 x1 6.9e-15 x2 1.9e-07 x3 2.7e-09 While coef(summary(m1))[,4] gives me: (Intercept) 3.0e-23 x1 5.7e-13 x2 2.6e-07 x3 1.7e-17 While the first one confirms my suspicion (-23 instead of -16), the latter one vary drastically (especially x3 from -09 to -17). Why is that? Can you show the complete code and output? Uwe Ligges Thank you! T. [[alternative HTML version deleted]] __** R-help@r-project.org mailing list https://stat.ethz.ch/mailman/**listinfo/r-helphttps://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/** posting-guide.html http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. [[alternative HTML version deleted]] __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
