Precision Estimates of AASHTO T265: Laboratory Determination of Moisture Content of Soils (2010)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

6 CHAPTER 3- INTERLABORATORY TEST RESULTS AND ANALYSIS 3.1 Test Data The moisture data were received for four aggregate-soil blends: coarse with clay (CC), coarse with silt (CS), fine with clay (FC), and fine with silt (FS). The moisture content data are provided in the tables in Appendices B thorough E. The empty cells in the tables indicate that the laboratory did not submit data and the shaded cells indicate that the data were considered as outliers and were eliminated from the analysis. The collected data are shown in Figures B-1, C-1, D-1, and E-1 of the appendices. In each graph, the middle point represents the median and the lower and upper bars represent the minimum and maximum data values, respectively. Appendices B through D also provide the graphical representation of the computed and critical h- and k- statistics (Graph B-2, C-2, D-2, and E-2), which are defined in ASTM E 691 for determining the outlier data. 3.2 Method of Analysis Test results of the ILS were analyzed for precision in accordance to ASTM E 691[2]. Prior to the analysis, any partial sets of data were eliminated by following the procedures described in E691 in determining repeatability (Sr) and reproducibility (SR) estimates of precision. Data exceeding critical h and k values were eliminated as described in Sections 3.3. Once identified for elimination, the same data were eliminated from any smaller subsets analyzed. 3.3 Analysis of Results 3.3.1 Coarse Blend with Clay The moisture content measurements of coarse aggregate with clay were received from 29 laboratories. The average and the repeatability and reproducibility standard deviations of the data were determined after eliminating the outlier data. The eliminated data are shown shaded in Table B-1 and are shown in Figure B-2 of Appendix B. Based on exceedance of h and k statistics, the results of two laboratories for the blend with 3 % moisture, the results of one laboratory for the blend with 5 % moisture, and the results of 4 laboratories for the blend with7 % moisture were eliminated from the analysis. All remaining data were re-analyzed according to E691 method to determine the Sr and SR precision estimates shown in Table 3-1. As indicated from the table, the average target values of 3 %, 5%, and 7 % were met very well by the average measured values of 3.02%, 4.98 %, and 6.89 %. It is also observed from the table that the variability of the water content measurement increased with increase in percentage of water.

7 Table 3-1: Summary of Statistics of % moisture content of coarse aggregate with clay (CC) Sample Type  # of  Labs  Target  %  Average  %  Sx  CV %  Repeatability  (Sr)  Reproducibility  (SR)  1s, %  d2, %  1s, %  d2s, %  Coarse Aggregate w/Clay  (3%)  27  3.0  3.02  0.06  1.9  0.042  0.1  0.07  0.2  Coarse Aggregate w/Clay  (5%)  28  5.0  4.98  0.11  2.3  0.044  0.1  0.12  0.3  Coarse Aggregate w/Clay  (7%)  25  7.0  6.89  0.26  3.8  0.060  0.2  0.27  0.8  3.3.2 Coarse Blend with Silt The moisture content measurements of coarse aggregate with silt (CS) were received from 30 laboratories. The average and the repeatability and reproducibility standard deviations of the data were determined after eliminating the outlier data. The eliminated data are shown shaded in Table C-1 and are shown in Figure C-2 of Appendix C. Based on exceedance of h and k statistics, the results from the following number of laboratories were eliminated from the analysis: two laboratories for blends with 3% moisture, one laboratory for the blends with 5% moisture, and one laboratory for the blend with7% moisture. All remaining data were re-analyzed according to E691 method to determine the Sr and SR precision estimates shown in Table 3-2. As indicated from the table, the average target values of 3 %, 5%, and 7 % were met relatively well by the average measured values of 3.03 %, 5.02 %, 6.60 %. Similar to the blend with clay, the variability of the moisture content measurements of the CS blends has increased with the increase in the level of moisture content. Table 3-2: Summary of Statistics of % moisture content of coarse blend with silt (CS) Sample Type  # of  Labs  Target  %  Average  %  Sx  CV %  Repeatability  (Sr)  Reproducibility  (SR)  1s, %  d2,  %  1s, %  d2s,  %  Coarse aggregate w/ Silt  (3%)  27  3.0  3.03  0.05  1.6  0.05  0.1  0.06  0.2  Coarse aggregate w/ Silt  (5%)  29  5.0  5.02  0.10  2.1  0.06  0.2  0.12  0.3  Coarse aggregate w/ Silt  (7%)  29  6.6  6.60  0.33  5.0  0.44  1.2  0.49  1.4  3.3.3 Fine Blend with Clay The moisture content measurements of fine aggregate with clay (FC) were received from 31 laboratories. The average and the repeatability and reproducibility statistics of the data were determined after eliminating the outlier data. The eliminated

8 data are shown shaded in Table D-1 and are shown in Figure D-2 of Appendix D. Based on exceedance of h and k statistics, the results from the following numbers of laboratories were eliminated from the analysis: one laboratory for the blend with 4% moisture, two laboratories for the blend with 6% moisture, and one laboratory for the blend with 8% moisture. All remaining data were re-analyzed according to E691 method to determine the statistics shown in Table 3-3. As indicated from the table, the average measured values of 4.04%, 5.92% are in very good agreement with the target values of 4% and 6%. However, for the blend with 8% moisture, the average measured value of 7.39 % was considerably below the expected value. In addition, the variability of the measurements for the blend with 8 % moisture as indicated from the standard deviations and coefficient of variation (CV %) was considerably larger than those of the blends with 4 % and 6 % moisture (CV of 8.5 % vs. 3.4%). It is speculated that this large variability is caused by higher probability of moisture loss during shipment and storage for the blends with above optimum moisture content than those with below optimum moisture content. When mixture is above the optimum, free moisture particles are available to evaporate and escape from microscopic pores of the bottles. While in mixtures below the optimum and at the optimum, moisture particles are adhered to the soil-aggregate particles. Table 3-3: Summary of Statistics of % moisture content of fine blend with clay (FC) Sample Type  # of  Labs  Target  %  Averag e %  Sx  CV %  Repeatability  (Sr)  Reproducibility  (SR)  1s, %  d2s, %  1s, %  d2s, %  Fine Aggregate w/ Clay  (4%)  30  4.0  4.04  0.14  3.4  0.18  0.5  0.20  0.6  Fine Aggregate w/ Clay  (6%)  29  6.0  5.92  0.20  3.4  0.17  0.5  0.25  0.7  Fine Aggregate w/ Clay  (8%)  30  8.0  7.39  0.63  8.5  0.73  2.0  0.87  2.4  3.3.4 Fine Blend with Silt The moisture content measurements of fine aggregate with silt (FS) were received from 31 laboratories. The average and the repeatability and reproducibility statistics of the data were determined after eliminating the outlier data. The eliminated data are shaded in Table E-1 and are shown in Figure E-2 of Appendix E. Based on exceedance of h and k statistics, the results from two laboratories for the blend with 4% moisture, one laboratory for the blend with 6% moisture, and one laboratory for the blend with 8% moisture were eliminated from the analysis. All remaining data were re-analyzed according to E691 method to determine the statistics shown in Table 3-4. Similar to the observation for the FC blends, the target values of 4 % and 6% for the FS blends were met very well by the average measured values of 3.97 % and 5.97 %. However, for the blend with above optimum moisture content, the average measured value of 7.69 % was considerably below the expected value of 8%. The variability of the data as indicated from standard deviations and coefficient of variation (CV %) was also considerably larger for the blend with 8 % moisture than the blends with 4 % and 6 % moisture (CV of

9 6.0 % vs. 2.9% and 2.7%). It is speculated that this larger variability is caused by the higher probability of moisture loss during shipment and storage of the blends with above optimum moisture content than those with below and at optimum moisture content. Table 3-4: Summary of Statistics of % moisture content of fine blend with silt (FS) Sample Type  # of  Labs  Target  %  Average  %  Sx  CV %  Repeatability  (Sr)  Reproducibility  (SR)  1s, %  d2s, %  1s, %  d2s, %  Fine Aggregate w/ silt  (4%)  29  4.0  3.97  0.11  2.9  0.17  0.5  0.18  0.5  Fine Aggregate w/ silt  (6%)  30  6.0  5.97  0.16  2.7  0.12  0.3  0.19  0.5  Fine Aggregate w/ silt  (8%)  30  8.0  7.69  0.46  6.0  0.60  1.7  0.68  1.9  3.4 Tests for Statistical Significance The tests of statistical significance were conducted to examine the significance of the differences in the average and standard deviations of the measurements. For each blend, a one sample t-test was performed to examine the significance of the difference between the average measured and expected percentage of moisture. In addition, an F-test on variance was performed to examine if the standard deviations of the measurements are different for different gradations, filler types (clay or silt), and moisture contents. The following section discusses the result of the statistical analysis. 3.4.1 Comparison of the Average Measured and Target Moisture Contents 3.4.1.1 Coarse Aggregate with Clay (CC) The results of the t-test for comparison of average and target moisture content of the coarse aggregate with clay for 1 % level of significance are provided in Table 3-5. A rejection probability (p) of smaller than 0.01 would indicate that the average measured value is significantly different from the target moisture level of the blend. As shown in Table 3-5, the p values are all above 0.01 indicating that the measured moisture contents of the coarse blend with clay were the same as the target moisture contents for that blend.

10 Table 3-5: Results of t-test for comparison of measured and target moisture content of CC blend Sample Type  Comparison  Computed t  Degrees  of  Freedom   Critical  t  Rejection  Prob. (P)  Decision  Coarse Aggregate w/Clay  (3%)  3.00% vs.  3.02%  2.140  26  2.779  0.042  Accept  Coarse Aggregate w/Clay  (5%)  5.00% vs.  4.98%  0.967  27  2.771  0.342  Accept  Coarse Aggregate w/Clay  (7%)  7.00% vs.  6.89%  2.022  24  2.797  0.055  Accept  3.4.1.2 Coarse Aggregate with Silt (CS) The results of a t-test for comparison of the measured and target moisture content of the coarse aggregate with silt for 1 % level of significance are provided in Table 3-6. A rejection probability (p) of smaller than 0.01 would indicate that the average measured value is significantly different than the target moisture level of the blend. As shown in Table 3-6, the p value for the comparison of the measured and target values of the CS blend with 3% moisture is slightly smaller than 0.01 indicating that the measured and target values are different. However, this decision is mainly due to the small standard deviation of the measurements and not the large difference between the average and target values. The p values corresponding to the blends with 5% and 7 % moisture indicate statistical agreement between the measured and target values. Table 3-6: Results of t-test for comparison of measured and target moisture content of CS blend Sample Type  Comparison   Computed  t  Degrees  of  Freedom   Critical t  Rejection  Prob. (P)  Decision  Coarse aggregate w/  Silt (3%)  3.00 % vs.  3.03 %  2.880  26  2.779  0.0079  Reject  Coarse aggregate w/  Silt (5%)  5.00 % vs.  5.02 %  0.783  28  2.763  0.4405  Accept  Coarse aggregate w/  Silt (7%)  7.00 % vs.  6.60 %  0.047  28  2.763  0.9626  Accept    3.4.1.3 Fine Aggregate with Clay (FC) The results of a t-test on average moisture content of the fine aggregate with clay for 1 % level of significance are provided in Table 3-7. A rejection probability (p) of smaller than 0.01 would indicate that the average measured value is significantly different than the target moisture level of the blend. As shown in Table 3-7, the p value for the comparison of the measured and target values of the FC blends with 3 % and 5% are larger than 0.01 indicating that the measured values are the same as the target values. However, the p value for the blend with 8% moisture is significantly smaller than 0.01

11 indicating that the measured and target values are not the same. As discussed previously, this significant difference between measured and target moisture content values might be due to loss of moisture during shipment of the mixtures with above optimum moisture content. Table 3-7: Results of t-test on comparison of measured and target moisture content of fine aggregate with clay Sample Type  Comparison   Computed t  Df  Critical t  P  Decision Fine Aggregate w/ Clay (4%)  4.00 % vs. 4.04 %  1.719  29  2.756  0.0963  Accept  Fine Aggregate w/ Clay (6%)  6.00 % vs. 5.92 %  2.002  28  2.763  0.0551  Accept  Fine Aggregate w/ Clay (8%)  8.00 % vs. 7.39 %  5.327  29  2.756  <0.0001  Reject  3.4.1.4 Fine Aggregate with Silt (FS) The results of a t-test on average moisture content of the fine aggregate with silt for 1 % level of significance are provided in Table 3-8. A rejection probability (p) of smaller than 0.01 would indicate that the average measured value is significantly different from the target moisture level of the blend. As shown in Table 3-8, the p value for the comparison of the measured and target values of the FS blends with 3 % and 5 % moisture are larger than 0.01 indicating that the measured values are the same as the target values. However, the p value for the blend with 8% moisture is significantly smaller than 0.01, indicating that the measured and target values are different. Similar to the previous reasoning, the significant difference between measured and target moisture contents might be due to the loss of moisture during shipment of the mixtures with above optimum moisture content. Table 3-8: Results of t-test on comparison of measured and target moisture content of FS blend Sample Type  Comparison   Computed t  Df  Critical t  P  Decision Fine Aggregate w/ silt (4%)  4.00 % vs. 3.97 %  1.481  28  2.763  0.1498  Accept  Fine Aggregate w/ silt (6%)  6.00 % vs. 5.97 %  1.138  29  2.756  0.2645  Accept  Fine Aggregate w/ silt (8%)  8.00 % vs. 7.69 %  3.627  29  2.756  0.0011  Reject  3.4.2 Comparison of the Variability of Moisture Content Measurement The preparation of the precision estimates for moisture content determination requires combining the standard deviations that are statistically similar. Statistical F-test on variances was performed to examine the significance of the difference between the variances. This was done in three steps. In the first step, the standard deviations of the clay and silt blends, for each gradation, will be compared at each moisture level. If the

12 standard deviations of the clay and silt blends, at a specific moisture level, are not significantly different, they would be combined. In the second comparison, the combined standard deviations at different moisture levels will be compared for each gradation. The standard deviations from various moisture levels that are not significantly different would be combined. In the third comparison, the combined standard deviations of different gradations will be statistically compared. If the standard deviations of the coarse and fine blends are not different they would be combined, otherwise they would be reported separately. The following sections provide the results of the statistical comparisons on variances. 3.4.2.1 Statistical Test on Standard Deviations of Various Blends A statistical F- test for comparison of the variances of measurements on coarse blend with clay and coarse blend with silt would indicate if the standard deviations from the two blends can be combined. The results of an F-test on comparison of the repeatability (Sr) and reproducibility (SR) standard deviations of the coarse blends, at each moisture level, are shown in Table 3-9. The comparison of the computed and critical F values on Sr or SR estimates at 1% level of significance indicates that the standard deviations are the same for clay and silt blends at 3 % and at 5% moisture level. However, the standard deviations of the clay and silt blends are significantly different at 7% moisture level. The small rejection probability values (<0.0001 and 0.002) corresponding to the comparisons of the variability of the blends with 7% moisture specifies the problem with the test samples having above optimum moisture content. Table 3-9: Results of F-test on comparison of variability of moisture content measurements of coarse aggregate with clay (CC) and coarse aggregate with silt (CS), critical F value correspond to 1% level of significance Comparison   Df  Cr.  F  Repeatability   Reproducibility  Standard  Deviations  Comp.  F  p  Decision  Standard  Deviations  Comp.  F  p  Decision  CC vs. Cs  3%  26 vs. 26  2.55  0.05 vs. 0.04  1.29  0.260  Accept  0.07 vs. 0.06  1.19  0.330  Accept  CC vs. Cs  5%  28 vs. 27  2.50  0.06 vs. 0.04  2.03  0.035  Accept  0.12 vs. 0.12  1.09  0.412  Accept  CC vs. Cs  7%  28 vs. 24  2.60  0.44 vs. 0.06  50.36  <0.0001  Reject  0.49 vs. 0.27  3.31  0.002  Reject  Table 3-10 provides the pooled repeatability and reproducibility standard deviations of the coarse blends with clay and silt. Since the standard deviations for the clay and silt at 3% moisture level were not significantly different they were combined. Similarly, the standard deviations for the coarse clay and coarse silt at 5 % moisture level were not significantly different and they were combined. However, the standard deviations for the coarse clay and coarse silt with 7 % moisture were significantly different and are presented separately.

13 Table 3-10: Pooled repeatability and reproducibility standard deviations of the clay and silt blends Blend Type  Moisture Content of  the Blend  Pooled Repeatability  Pooled  Reproducibility  Coarse blend (Clay & Silt)  3%  0.04  0.06  Coarse blend (Clay & Silt)  5%  0.05  0.12  Coarse blend (Clay)  7%  0.06  0.27  Coarse blend (Silt)  7%  0.44  0.49  Table 3-11 provides the results of statistical F test on comparison of the variances of measurements on fine blend with clay and fine blend with silt at various moisture levels. The comparison of the computed and critical F values for 1% level of significance indicates that there is no significant difference between either Sr or SR estimates for moisture content of the clay and silt blends at any moisture content. Therefore, the standard deviations of the fine silt and fine clay blends could be pooled at each moisture level. Table 3-11: Results of F-test on comparison of variability of moisture content measurements of FC and FS blends Comparison   Df  Cr.  F  Repeatability   Reproducibility  Standard  Deviation  Comp.  F  p  Decision  Standard  Deviation  Comp.  F  p  Decision  FC vs. FS (4%)  29 vs. 28  2.45  0.18 vs. 0.17  1.08  0.42  Accept  0.20 vs. 0.18  1.22  0.30  Accept  FC vs. FS (6%)  28 vs.29  2.44  0.17 vs. 0.12  1.97  0.04  Accept  0.25 vs. 0.19  1.65  0.09  Accept  FC vs. FS (8%)  29 vs. 29  2.42  0.73 vs. 0.60  1.46  0.16  Accept  0.87 vs. 0.68  1.64  0.09  Accept  Table 3-12 provides the pooled standard deviations of FC and FS blends at different moisture contents. As shown in the table, the pooled standard deviation of fine blend at 8% moisture level is considerably larger than the standard deviations of the blend at other moisture levels indicating the problem with the blends at above optimum moisture content. The statistical significance of the difference in standards deviations of various moisture levels will be examined in next Section. Table 3-12: Pooled repeatability and reproducibility standard deviations of the clay and silt blends Blend Type  Moisture Content of  the Blend  Pooled Repeatability  Pooled  Reproducibility  Fine blend (Clay & silt)  4%  0.17  0.19  Fine blend (Clay & silt)  6%  0.15  0.22  Fine blend (Clay & silt)  8%  0.67  0.78 

14 3.4.2.2 Statistical Test on Standard Deviations of Various Moisture Levels The results of F-test at 1% level of significance for comparison of repeatability and reproducibility statistics of the coarse blends are presented in Table 3-13. As indicated from Column 5, Table 3-13, the rejection probabilities (p) corresponding to the repeatability of the blends with 3 % and 5% moisture were not significantly different from each other and from the repeatability of CC blend at 7% moisture. Therefore, they can be combined. However, the standard deviation of the CS blend at 7% moisture content was significantly larger than those of other coarse blends and could not be combined. The rejection probabilities, in Column 7 of Table 3-13, from comparison of the reproducibility standard deviations indicates that the standard deviations corresponding to different moisture levels were all significantly different from each other (p<0.0001) and could not be combined. Table 3-13: Results of F-test on comparison of variability of moisture content measurements of coarse blends at various moisture contents Compare  Df  Critical  F  Computed F  (Repeatability)  p  (Repeatability)  Computed F  (Reproducibility)  p  (Reproducibility)  5% (CC & CS) vs.   3% (CC & CS)  56 & 53  1.90  1.44  0.09  3.37  <0.0001  7% (CC) vs. 3%  (CC & CS)  24 & 53  2.16  1.91  0.03  17.55  <0.0001  7% (CC) vs. 5%  (CC & CS)  24 &56  2.14  1.33  0.19  5.21  <0.0001  7% (CS) vs. 3%  (CC & CS)  28 & 53  2.29  96.01  <0.0001  58.36  <0.0001  7% (CS) vs. 5%  (CC & CS)  28 & 56  2.15  66.78  <0.0001  17.33  <0.0001  The results of F-test at 1% level of significance for comparison of repeatability and reproducibility statistics of the fine blends are presented in Table 3-14. As indicated from the p values (Column 5), the standard deviations corresponding to 4% and 6% moisture were not significantly different from each other and could be combined. However, the standard deviations corresponding to 8% moisture were significantly different from the blends with 4% and 5% moisture and could not be combined.

15 Table 3-14: Results of F-test on comparison of variability of moisture content measurements at various moisture contents, fine aggregate with clay (FC), and fine aggregate with silt (FS) Compare  Df   Critical  F  Computed F  (Repeatability)  p  (Repeatability)  Computed F  (Reproducibility)  p  (Reproducibility)  4% (FC & FS) vs.  6%(FC & FS)  58 & 58  1.86  1.38  0.11  1.35  0.13  8% (FC & FS) vs.  4%(FC & FS)  59 & 58  1.87  15.01  <0.0001  16.86  <0.0001  8% (FC & FS) vs.  6% (FC & FS)  59 & 58  1.87  20.69  <0.0001  12.50  <0.0001  3.4.3 Combined Standard Deviations of Various Moisture Levels To prepare the precision estimates, the standard deviations were combined based on the rationality of the values and based on the significance of their differences. For the coarse blends, the large standard deviation of the coarse blends at 7% moisture appeared to be suspect. Therefore, it was judged that both repeatability and reproducibility standard deviations of CS blend at 7% to be eliminated from the precision estimate calculation. Similarly, for the fine blends, both repeatability and reproducibility standard deviations of the blend with 8% moisture seemed unreasonably high and were eliminated from the analysis. The remaining standards deviations were pooled to compute separate standard deviations for the coarse and fine blends as provided in Table 3-15. Table 3-15: Computed standard deviations for water content measurement of coarse and fine blends Blend Type  Repeatability  Standard deviation,  %  Reproducibility  Standard deviation,  %  Coarse blend    0.05  0.12  Fine blend  0.16  0.21  To examine if the standard deviations can be further combined, an F-test was conducted to examine the significance of the difference between variability of the coarse and fine blends. The results of statistical F test at 1 % level of significance are provided in Table 3-16. As seen from the table, the comparison indicates significant difference between the standard deviations of the coarse and fine blends. Therefore, the variability of the blends cannot be further combined and would be presented separately.

16 Table 3-16: Results of F test for comparison of standard deviations of water content measurements of coarse and fine blends Compare  Degrees  of  Freedom  Critical F  Computed  F(Sr)  Rejection  Probability  (SR)  Computed  F(SR)  Rejection  Probability (SR)  Coarse &  Fine  117 & 110  1.58  8.85  <0.0001  2.92  <0.0001  3.4.4 Precision Estimates of AASHTO T265 Table 3-17 provides the precision estimates for water content determination based on the results of the ILS conducted in this study. The standard deviations corresponding to coarse and fine blends in Table 3-15 were used to compute the allowable differences between two water content measurements. A proposed precision statement for NCHRP T265, based on the precision estimates in Table 3-17 is provided in Appendix F. Table 3-17: Pooled standard deviations of the blends with various moisture contents Material and Type Index  Standard deviations  (1s)  Acceptable Range  of Two Results  (d2s)  Single‐Operator Precision:  Coarse blend  Fine blend  0.05  0.16  0.14  0.46  Multilaboratory Precision:  Coarse blend  Fine blend    0.12  0.21    0.33  0.58