Project Summaries

09-631  Project Manager: D. C. Jones


William C. Bridges, Clemson University

This project involves the study of estimating cotton yarn quality from fiber quality. There are two methods of measuring fiber variables to estimate yarn quality, HVI and AFIS. Using regression modeling techniques with random year, location, and variety effects, we determined that AFIS predicts a more precise estimate of spun cotton quality than HVI. Although AFIS gives better estimates, it is also more expensive than HVI, so the second part of the project develops a relationship between the estimated quality and cost. With this relationship, we determined conditions in which it should be worth investing in the more expensive method.

The United States Department of Agriculture Marketing Services currently uses the High Volume Instrument (HVI) to obtain fiber quality measures of all cotton bales. The Advanced Fiber Information System (AFIS) can also be used to obtain fiber quality measures, but AFIS is slower and three times more expensive than HVI ($15 per bale versus $5 per bale). Previous research, along with this project, has indicated that there is an important predictive relationship between primary measures of spun yarn quality and the cotton fiber quality measurements from both HVI and AFIS. AFIS has also shown to have relationships resulting in a more precise estimation of spun yarn quality. However, as mentioned earlier, AFIS is also more expensive compared to HVI.

The goal of this project was to carefully analyze a data set containing both HVI and AFIS measures and actual spun yarn quality. This involved two stages. The first stage involved regression model building with a combination of fiber quality measures and random effects. The second stage was to determine if the difference in estimation precision between HVI and AFIS was worth the extra cost and time.  The data set was provided by the project manager. The set involves a sample of 97 cotton bales, and for each bale there are measured values of the nine HVI variables, 19 AFIS variables, and several measures of spun cotton quality.  Adjusted break factor [ABF] is the one we focus on in this report

We found that AFIS provided slightly more precise estimates of ABF than HVI. When considering models with significant variables and including effect model terms as random effects, AFIS had RMSE=106.5051 and R2 = .847669 while HVI had RMSE=108.637 and R2 =.845357. The next stage was to evaluate whether the investment in AFIS was worth the slight increase in precision over HVI. The obvious answer appears to be no since the increase is so small, but we decided to complete the exercise of using cost in the precision decision.

The cost of measuring the actual ABF for cotton is ~$400, while the cost of measuring using HVI and AFIS are ~$5 and $15, respectively. An important first step was to determine if the estimation of ABF was cost effective overall. Based on the general concept of cost functions (Zellner 1968), we attempted to develop an approach to determine if the savings in cost is 'worth' the loss in precision. We started by creating a model relating the quality measurments and the dollar value of cotton bales. We assumed a simple, linear relationship.  As long as costs are less than $385 ($400 - $15) and $395 ($400 - $5), respectively for AFIS and HVI, estimation is cost effective.  When developing and building the model, we noticed a significant difference between effect terms when they were considered fixed effects versus random effects. We saw that many model effect terms and interactions were considered significant if Loc, Variety, and Year were treated as fixed effects, while none of these effects were considered significant when these were treated as random effects. We also note that omitting the Loc, Variety, and Year from the model completely had a large impact on the RMSE and R2. However, since Loc, Variety, and Year were subsamples of all locations, varieties, and years, it is best to treat them as random effects.


Project Year: 2012

Search 2012 Projects:

▸ Alabama
▸ Arizona
▸ Arkansas
▸ California
▸ Florida
▸ Georgia
▸ Louisiana
▸ Mississippi
▸ Missouri
▸ New Mexico
▸ North Carolina
▸ Oklahoma
▸ South Carolina
▸ Tennessee
▸ Texas
▸ Virginia

▸ Cotton Incorporated Fellow
▸ Cottonseed
▸ Crop Improvement
▸ Farm-to-Mill
▸ Production Efficiency
▸ Sustainable Cotton
▸ Variety Improvement

Project Manager

▸ All Project Nos.

Share This: