Experimental Design in Agriculture CSS 590 Final Exam, Winter, 2015 Name_______KEY___________ Part 1. Short answer, multiple choice, brief discussion questions 1) An experiment was conducted to determine the effects of two soil amendments on dry weight of four cultivars of a perennial grass species. A control treatment (no amendment) was also applied to each cultivar. The treatments were arranged in a split-plot design with soil amendment as the main plot and cultivar as the subplot. The experiment was replicated in four complete blocks. Complete the ANOVA (fill in the shaded areas): 11 pts Source df SS Total Block 47 3 2023 294 98 Amendment 2 1420 710 Error a 6 48 8 Cultivar 3 69 23 5.75 Amendment x Cultivar 6 84 14 3.5 27 108 4 Error b 6 pts MS F 88.75 a) Using the F table in the back of this exam, what are your conclusions regarding the effects of soil amendment and cultivar treatments on dry weight of this grass species? Justify your answer. The Amendment x Cultivar interactions are significant (3.5 is greater than Fcritical = 2.46). The main effects of amendments is highly significant (88.75>>5.14). The main effect of cultivars is significant (5.75>2.96). However, the main effects should be interpreted with caution due to the presence of the interactions. The effect of the soil amendment depends on the cultivar. 6 pts b) How would you report the results? Calculate the appropriate standard error(s) for the means. You would need to report the means for each combination of cultivar and soil amendment. The standard error of the mean would be se MSE r 4 1 4 The relatively large effects of the amendments suggests that it might be possible to draw general conclusions about them in spite of the interactions with cultivars. You would want to explore this further by graphing the means and trying to understand the nature of the interaction. 1 Question #1 cont’d. 8 pts c) The researcher would like to obtain additional harvests from the same plots for several years. What approach would you recommend for conducting a combined analysis of the data across years? Explain the rationale for your choice. Use a repeated measures analysis when you are taking repeated observations from the same experimental units over time. There is likely to be some correlation in errors from one harvest to the next, and the repeated analysis adjusts for that. In order for a split-plot to be valid, you have to be able to assume that correlations in errors among the sub-plot observations are equal. That is not likely to be the case for repeated measures in time, because observations that are made at close time intervals are likely to be more similar than those that are taken at distant time intervals. With a repeated measures analysis, patterns in the covariance structure can be taken into account. An autoregressive covariance structure is often appropriate for repeated measures in time. 2) An experiment was conducted to determine the optimum time to apply a plant growth regulator to reduce lodging in oats. The growth regulator was applied at three growth stages (ZGS22, ZGS26, and ZGS30). A control treatment (no growth regulator) was also included. The experiment was conducted for three years to see if the optimum application time and effect of the growth regulator are consistent across a range of environmental conditions. The experimental design was a randomized complete block design with four replications. Source df Mean Square Expected Mean Square Year 2 MS1 σ2e + 4σ2Rep(Year) + 16σ2Year Rep(Year) 6 MS2 σ2e + 4σ2Rep(Year) Growth Stage 3 MS3 σ2e + 4σ2Year*GS + 12Ө2GS Year*Growth Stage 6 MS4 σ2e + 4σ2 Year*GS Error 27 MS5 σ2e Based on the Expected Mean Squares given in the table above, what would be appropriate ratio of Mean Squares to use to calculate an F value 6 pts a) To determine if there are differences among the Growth Stage Treatments? MS3/MS4 6 pts b) To determine if there are differences among the Years? MS1/MS2 2 3) You are a food technologist and you have developed three new methods for making orange juice. You wish to evaluate consumer acceptance of juice made with these new methods. You also want to compare the new juice products to two types of orange juice that are currently on the market (fresh squeezed brand X and frozen brand Y). You suspect that different age groups may have different preferences. Ten panelists in each of three age groups are identified to participate in the study (10 young, 10 middle-aged, and 10 elderly adults). Each of the panelist will be given the five types of orange juice in random order and asked to rate them for various quality parameters on a questionnaire that you have provided. Answer the two questions below regarding the linear model for this experiment. “Products” refer to the five types of orange juice. Question 1 – circle the best answer 6 pts a) Panelists and Age groups are cross-classified b) Panelists are nested in Products c) Age groups and Products are cross-classified d) There are no nested effects in the model Question 2 – circle the best answer 6 pts a) Panelists are fixed effects b) Products are fixed effects c) Age is the only fixed effect in the model d) All effects in the model are random 4) A fellow graduate student is planning an experiment, and seems to think that more complex experimental designs are better than simple designs. How would you convince him that it is best to use the simplest possible design that will meet the objectives of the experiment? Include at least three reasons that you would give him to justify your position. 9 pts - - - Any additional blocking factors will impose more constraints on your randomization and will remove degrees of freedom from error, thereby reducing the power of the significance tests. Blocking will only be beneficial if it is effective in reducing experimental error. Designs with multiple plot sizes may be necessary in particular circumstances, but complicate the statistical analysis and mean comparison tests, and generally result in fewer degrees of freedom for error (because there are several error terms). Missing plots are more problematic with complex designs than with simple designs such as a CRD. Three-way and higher order factorials can become very difficult to interpret, particularly if interactions are significant. Greater complexity in planning, implementation, data collection and analysis provides more opportunities for mistakes. There must be a clear benefit to justify the use of a more complex design. 3 5) You are studying the effect of three cultivation methods on fresh weight of spinach. You decide to use a Randomized Block Design with 5 Blocks. This is your first experience collecting data of this sort, and you are not sure about the appropriate plot size needed to obtain an acceptable level of precision. For a preliminary analysis, you collect samples from two quadrats in each plot, and you then ask your assistant to enter the data and calculate the ANOVA. He provides you with the output below: The GLM Procedure Dependent Variable: weight Source DF Sum of Squares Mean Square F Value Pr > F Model 6 16.35733333 2.72622222 Error 23 8.92266667 0.38794203 Corrected Total 29 25.28000000 7.03 0.0002 R-Square Coeff Var Root MSE weight Mean 0.647046 11.12232 0.622850 5.600000 Source DF Type III SS Mean Square F Value Pr > F 9 pts block 4 4.16333333 1.04083333 2.68 0.0570 method 2 12.19400000 6.09700000 15.72 <.0001 Looking at the degrees of freedom and F ratios, you realize that the analysis has not been done correctly. How would you explain the mistake to your assistant? What should be done to obtain a correct analysis? The residual in his analysis includes variation among plots treated alike (true experimental error) and variation among quadrats within each plot (sampling error). Pooling the two sources of variation together will give you too many degrees of freedom in the error term, and will probably provide an estimate of error that is too small, thereby inflating the Type I error rate. One approach is to calculate the means for each plot and perform ANOVA on the means. The other option is to keep the data for individual quadrats in the data set, but specify that the appropriate error for testing methods is the block*method interaction. In an RBD, this term represents the error among experimental units to which the treatments were randomly applied. An analysis including the individual quadrats would have the benefit of providing an estimate of sampling error, which could provide insights about the plot size that would be needed to meet experimental objectives. Correct ANOVA including subsamples: Source Total Block Method Block*Method Sampling Error 4 df 29 4 2 8 15 Part 2. Experimental Design Question You are planning an experiment to estimate yield losses due to an insect pest on an oilseed crop. The current practice is to spray with an insecticide one time during the cropping season. You would like to know if an additional spray one month later could provide a higher level of control and reduce yield losses. To avoid driving over the crop, the insecticide is generally applied to your research plots by driving through an alley that separates the plots. The boom on the sprayer can be adjusted to spray insecticide on one or both sides of the tractor. When fully extended on both sides, the width of the boom is 25 feet. The movement of the tractor with sprayer is perpendicular to the direction of planting. The alley must be at least 4 feet wide to avoid driving over the ends of the plots. You expect that the width of spacing between rows can affect the damage due to the insect and the capacity of the plants to compensate for the insect damage. Adjusting the row width on the planter is time consuming and must be done before or after planting any particular pass of the planter through the field. The planter can be adjusted to a 6”, 9” or 12” row spacing and can plant individual plots that are 20 feet long. The planter is 6-feet wide. Design an experiment that would meet your objectives 6 pts 1) What type of experimental design will you use? Justify your choice. Indicate any basic assumptions that you have made. I will use a strip plot design, due to the fact that both the insecticide and the plant spacing treatments are best applied in a single pass through the field (i.e. it is difficult to change treatments in the middle of a pass). We have to assume that the insect pressure is uniform throughout the field, and that spraying one plot won’t affect other insecticide treatments nearby. 5 pts 2) List the treatments of the experiment. Be sure to include any necessary controls. Explain why you have chosen this particular set of treatments. Insecticide – 1) no spray 2) one spray at the recommended time 3) two sprays – recommended time plus a later spray Plant spacing - 6”, 9” and 12” The no spray treatment is needed as a control so that we know the extent of insect damage in the field when insects are not controlled, and to see if spraying is effective. This will be a complete 3 x 3 factorial. Due to the limited error degrees of freedom for a strip-plot experiment and the large errors associated with insect counts, I will use 5 replications. A randomized complete block design will be used to facilitate the application of the treatments (and there is really no practical option for using a CRD because the treatment factors are applied in perpendicular strips). 5 8 pts 3) Break out the ANOVA in terms of Sources of Variation and degrees of freedom. Indicate the Mean Squares that would be used to calculate the F ratios for testing the treatment main effects and interactions for the design that you have chosen. Source Total Block Insecticide Block * insecticide (error a) Spacing Block * spacing (error b) Insecticide * spacing Residual df 44 4 2 8 2 8 4 16 4) Draw a diagram to indicate the field layout. For one replication, show how the treatments will be randomized and assigned to experimental units. N 8 pts 6" 20 ft BLOCK 1 9" 12" one spray 6" BLOCK 2 12" 9" no spray 5 ft alley no spray two sprays 5 ft alley two sprays one spray An additional 3 Blocks would also be included. The planter would drive in the north –south direction. The sprayer would drive down the alleys (eastwest), spraying on one side at a time. Because the boom extends for 10 ft over the plots, two passes would be needed (up one alley and down the next) to spray the entire 20-ft plot. 6 ft The center of each plot would be harvested to reduce border effects due to differences in plant density in neighboring plots. The ends of the plots adjacent to the alleys could also be trimmed back before harvest. 6 F Distribution 5% Points Denominator Numerator df 1 2 3 4 5 6 7 1 161.45 199.5 215.71 224.58 230.16 233.99 236.77 2 18.51 19.00 19.16 19.25 19.30 19.33 19.36 3 10.13 9.55 9.28 9.12 9.01 8.94 8.89 4 7.71 6.94 6.59 6.39 6.26 6.16 6.08 5 6.61 5.79 5.41 5.19 5.05 4.95 5.88 6 5.99 5.14 4.76 4.53 4.39 4.28 4.21 7 5.59 4.74 4.35 4.12 3.97 3.87 3.79 8 5.32 4.46 4.07 3.84 3.69 3.58 3.50 9 5.12 4.26 3.86 3.63 3.48 3.37 3.29 10 4.96 4.10 3.71 3.48 3.32 3.22 3.13 11 4.84 3.98 3.59 3.36 3.20 3.09 3.01 12 4.75 3.88 3.49 3.26 3.10 3.00 2.91 13 4.67 3.80 3.41 3.18 3.02 2.92 2.83 14 4.60 3.74 3.34 3.11 2.96 2.85 2.76 15 4.54 3.68 3.29 3.06 2.90 2.79 2.71 16 4.49 3.63 3.24 3.01 2.85 2.74 2.66 17 4.45 3.59 3.20 2.96 2.81 2.70 2.61 18 4.41 3.55 3.16 2.93 2.77 2.66 2.58 19 4.38 3.52 3.13 2.90 2.74 2.63 2.54 20 4.35 3.49 3.10 2.87 2.71 2.60 2.51 21 4.32 3.47 3.07 2.84 2.68 2.57 2.49 22 4.30 3.44 3.05 2.82 2.66 2.55 2.46 23 4.28 3.42 3.03 2.80 2.64 2.53 2.44 24 4.26 3.40 3.00 2.78 2.62 2.51 2.42 25 4.24 3.38 2.99 2.76 2.60 2.49 2.40 26 4.23 3.37 2.98 2.74 2.59 2.47 2.39 27 4.21 3.35 2.96 2.73 2.57 2.46 2.37 28 4.20 3.34 2.95 2.71 2.56 2.45 2.36 29 4.18 3.33 2.93 2.70 2.55 2.43 2.35 30 4.17 3.32 2.92 2.69 2.53 2.42 2.33 7 Student's t Distribution (2-tailed probability) df 0.40 0.05 0.01 1 1.376 12.706 63.667 2 1.061 4.303 9.925 3 0.978 3.182 5.841 4 0.941 2.776 4.604 5 0.920 2.571 4.032 6 0.906 2.447 3.707 7 0.896 2.365 3.499 8 0.889 2.306 3.355 9 0.883 2.262 3.250 10 0.879 2.228 3.169 11 0.876 2.201 3.106 12 0.873 2.179 3.055 13 0.870 2.160 3.012 14 0.868 2.145 2.977 15 0.866 2.131 2.947 16 0.865 2.120 2.921 17 0.863 2.110 2.898 18 0.862 2.101 2.878 19 0.861 2.093 2.861 20 0.860 2.086 2.845 21 0.859 2.080 2.831 22 0.858 2.074 2.819 23 0.858 2.069 2.807 24 0.857 2.064 2.797 25 0.856 2.060 2.787 26 0.856 2.056 2.779 27 0.855 2.052 2.771 28 0.855 2.048 2.763 29 0.854 2.045 2.756 30 0.854 2.042 2.750