Enology Notes

Enology Notes #19 May 7, 2001

To: Virginia Vintners and Prospective Vintners

From: Bruce Zoecklein

Subject: Laboratory Methods Validation Program, Precision and Accuracy in the Winery Laboratory

Laboratory Methods Validation Program. The Enology-Grape Chemistry Group is starting a methods validation program. The initial phase will involve the analysis of pH, titratable acidity and alcohol. If you are interested, respond by email. You will be sent a standard with instructions. Following your analysis you will email your results and then receive a note back providing the actual pH, TA and percent alcohol by volume of the sample. In subsequent mailings you will be offered the opportunity to receive standards for the Formol N test, sugar analysis, volatile acidity and physical stability tests. We hope this review will help establish confidence in your wine lab analysis. For additional information see below.

Precision and Accuracy in the Winery Laboratory. (From a presentation given by Bob Whiton, Research Scientist, Enology-Grape Chemistry Group). Analytical data on wine and juice are of no use to the vintner unless he has some idea of how reliably the data represent the actual condition of the product. The quality of analytical data is a function of proper sampling, appropriate method selection, and careful execution. With practice, attention to detail, and the right method, the winery analyst can generate reliable and useful data.

In order to evaluate your analytical data, you need some measures of data quality. The most useful measures are precision and accuracy. Precision is the measure of reproducibility, or how close the results of replicate analyses are. Accuracy is the measure of how close the test results are to the true value. Precision and accuracy are two distinctly different concepts. A set of data can be precise without being accurate. Data can also be accurate without being very precise. It is important to remember the difference.

Precision is measured by performing replicate analyses on a sample and examining the spread of the data. An example of a set of acidity titrations performed by four students is shown in Table 1. The precision of each set of numbers is expressed by a statistical measure called standard deviation, which is calculated from the formula:

where is the average (or mean) of the results, x is an individual result, and n is the number or analyses (most spreadsheet programs and scientific calculators will perform this calculation for you). The standard deviation is often expressed as a percentage of the average result, which is referred to as relative standard deviation (RSD). If we look at the results for the four students, we see that students A and D reported very precise results, with RSDs less than 1%, while students B and C have not done quite so well.

Table 1. Results of titrations performed by four students.






Results (g/L):































Std. Deviation










Accuracy is determined by analyzing samples of known composition and calculating the deviation from the known value. In the case of our students, the sample was a 5.0 g/L solution of tartaric acid. We can see that students B and D have come quite close to that value, reporting very accurate results. Students A and C were off by about 3%. Overall, student D has generated the best data, being both precise and accurate.

Having calculated the precision and accuracy of the analytical results, we can now look at possible reasons for the differences in data quality. Precision is primarily affected by random errors. These include errors in measuring out a sample, errors in reading a buret, and failure to titrate to the exact endpoint each time. These random errors can be reduced by practice and by using the proper equipment. For example, you can measure 1 mL of sodium hydroxide solution more precisely with a 10 mL buret than with a 100 mL buret. Use of a pH meter instead of an indicator for endpoint determination generally produces more precise results; a meter reading to 0.01 units is better than one reading to 0.1 units. The precision of an analysis is also a function of the method being used. For example, determination of titratable acidity involves one sample measurement, which can be done very precisely with a pipet, and a titration, and can be performed with an RSD of 2% or less. On the other hand, a nitrogen determination by the Formol method involves an initial sample measurement, a dilution, another sample measurement, a formaldehyde addition, and finally a titration. Each step introduces some error, and since there are more steps than in the TA determination, the RSD will be higher, perhaps up to 3%. The students in the January short course demonstrated this very nicely, getting equally accurate results for TA and Formol, but with less precise results for the Formol determination.

Accuracy is affected by systematic errors, also known as bias. Bias can come from operator errors, such as failure to standardize a sodium hydroxide solution or not purging the carbon dioxide from a sample prior to a TA determination. Failure to standardize the sodium hydroxide is the most likely explanation of the results from students A and C in our example. Bias can also be due to instrumental error, such as an inaccurate pH meter or buret. Finally, bias can be caused by limitations of the method,such as interferences from other compounds in the sample. The analyst can limit these sources of analytical bias by careful method selection, and by following all of the steps in the method faithfully.

One more important source of analysis inaccuracy is the sample itself. Unless care is taken in the sampling process, a sample may be collected which is not representative of the entire lot of juice or wine. For example, a sample taken from the top of a tank may differ from one taken from the racking port. Sample packaging and storage may also adversely affect the analysis results, possibly through loss of volatile constituents or fermentation in the sample container. If the sample is flawed, then the best analysis will generate no useful information.

If these considerations are kept in mind, any winery laboratory should be able to generate useful analytical data.