You are on page 1of 3

Laboratory Experiment

pubs.acs.org/jchemeduc

Measuring the Gas Constant R: Propagation of Uncertainty and


Statistics
Robert J. Olsen*,† and Simeen Sattar‡

School of Natural Sciences and Mathematics, The Richard Stockton College of New Jersey, Galloway, New Jersey 08205-9441,
United States

Physics Program, Bard College, Annandale-on-Hudson, New York 12504-5000, United States
*
S Supporting Information

ABSTRACT: Determining the gas constant R by measuring the properties of hydrogen


gas collected in a gas buret is well suited for comparing two approaches to uncertainty
analysis using a single data set. The brevity of the experiment permits multiple
determinations, allowing for statistical evaluation of the standard uncertainty uR within a
laboratory period, while calculating R from the several measured quantities (pressure,
volume, etc.) invokes a number of different methods of estimating the individual
uncertainties involved in the calculation of the combined standard uncertainty uc(R). The
two uncertainties, as obtained by several classes of upper-level college students, are
typically within a factor of two or three. In the course of the data analysis, students
interpret a q-q plot, identify outliers using Dixon’s test, and apply the F test to the ratio of
the two variances. Thus, the experiment gives students experience with a variety of
methods for evaluating data they have gathered themselves. The Supporting Information
contains an Excel workbook containing student data and all the above-mentioned
calculations.
KEYWORDS: Upper-Division Undergraduate, Physical Chemistry, Laboratory Instruction, Computer-Based Learning,
Inquiry-Based/Discovery Learning, Hands-On Learning/Manipulatives, Gases, Student-Centered Learning

C ritical evaluation of experimental data and quantification


of uncertainty are fundamental to the practice of science.1
Owing to their importance, these concepts and skills are
quick to execute. A small class can produce a data set that is
large enough for statistical analysis if each student performs two
or three trials (each trial requires about 30 min). Depending on
developed from the start of the college chemistry curriculum in the length of a laboratory period and the extent to which the
general chemistry and reach their maximal sophistication in laboratory and classroom components of the course are
analytical and physical chemistry. Of the two main approaches integrated, students can analyze the data in the remainder of
to estimating uncertainty in a derived result, statistical analysis the laboratory period, in a subsequent classroom period, or as
and propagation of individual uncertainties, the first is given part of the process of writing the report. Through doing the
more attention in analytical chemistry, whereas the second falls data analysis, students learn how to propagate uncertainty, a
to the physical chemistry laboratory. Physical chemistry technique whose importance is affirmed by its inclusion in all
laboratory textbooks discuss both approaches in considerable current physical chemistry laboratory textbooks and its regular
detail.2−4 Coming at the beginning of the course, these appearance in this Journal.7−9 Of most significance, students
treatments can overwhelm students. When should each method produce their own experimental evidence in support of the
be applied? What are their advantages? How do they compare? validity of propagation of uncertainty. They extend their
Starting from a common general chemistry experiment5,6 in knowledge of statistics through use of Dixon’s test and the F
which the gas constant R is determined, we have developed an test, both of which arise naturally from the data analysis and
experiment for the physical chemistry laboratory that allows both of which are standard topics in physical chemistry
students to focus on these two approaches to uncertainty laboratory textbooks. This experiment can be used to introduce
analysis and answer these questions. Taken together, data students to a more sophisticated language of statistics and data
collection and analysis requires between three and four hours. analysis, with the instructor having considerable flexibility in

■ EXPERIMENTAL OVERVIEW
The gas constant depends on several measured quantities
determining the appropriate level of detail. Likewise, the
instructor has considerable flexibility in choosing how the
propagation of uncertainty is carried out: students may use
whose uncertainties must be estimated. Propagating these analytical derivatives,2−4 numerical derivatives,7−9 or a Monte
uncertainties produces an estimate of the uncertainty in R. The Carlo approach.9
laboratory component involves no new physical concepts or
unfamiliar instrumentation, so it is conceptually simple and
© XXXX American Chemical Society and
Division of Chemical Education, Inc. A dx.doi.org/10.1021/ed3005374 | J. Chem. Educ. XXXX, XXX, XXX−XXX
Journal of Chemical Education


Laboratory Experiment

EXPERIMENTAL PROCEDURE test, both used subsequently. A procedure for making the plot
This experiment has been tested by five classes of physical and a spreadsheet template are included with the Supporting
chemistry students; the refined procedure described here has Information.
been tested by a group of upper-level college chemistry and In the next class meeting, during which students have access
physics majors. to computers, potential outliers are first identified using the q-q
A small volume of hydrochloric acid solution is poured into a plot. Of the numerous methods for rejecting outliers, Dixon’s
gas buret. The remainder of the buret is filled with distilled test was selected because students might be familiar with it in
water. A strip of magnesium is suspended by a thread a few its simplest form (the Q test) and because it is conceptually
centimeters below the surface of the water. The buret is then easy to understand and apply. The critically reviewed data set is
inverted in a large beaker of water, bringing the denser posted and is the single data set from which all students work.
hydrochloric acid solution to the top. The acid flows Students assess their data in two ways. For the statistical
downward, reaches the magnesium, and reacts with it, analysis, they calculate the mean (R̅ ) and its standard
generating hydrogen gas, which collects at the top of the uncertainty uR, which is Sm, the standard deviation of the
buret. After the reaction is complete, the following data are mean. This result is reported in the form R̅ ± uR. Next, students
recorded: t, the Celsius temperature of the air near the beaker; calculate the combined standard uncertainty uc(R) for their
tsoln, the Celsius temperature of the stirred solution in the own trials and report their individual results as R ± uc(R). The
beaker; V, the volume of the gas; pdiff, the height difference instructor posts the values of uc(R) for the entire data set so
between the levels of solution in the buret and beaker; and patm, that students can check their calculations. Students determine
the barometric pressure. whether the accepted value of R lies in these intervals;


discrepancies between the observed values of R and its accepted
HAZARDS value lead to consideration of sources of systematic error. The
A 6 M HCl solution is corrosive and can be dangerous to the uncertainty analysis reveals that the uncertainties in the mass of
eyes and skin; eye protection and gloves should be worn. Spills Mg and volume of gas in the buret are the dominant
should be neutralized with sodium bicarbonate. The resulting contributions to uc(R).
sodium chloride and excess sodium bicarbonate can be In the final stage of the analysis, students first compare uR
disposed of in the sink. The solution that remains at the end and uc(R). The two measures of uncertainty are of the same
of the experiment (<0.09 M HCl, <0.006 M MgCl2) can be order of magnitude (typically within a factor of 3). As a next
disposed of in the sink. Unused magnesium can be disposed of step, students compare the relative uncertainties, uR/R̅ and
in the trash. uc(R)/R; these are typically less than 10−3, indicating that any


difference in the uncertainties is of limited practical
UNCERTAINTY ANALYSIS significance.11
For a quantitative comparison, an F test12 may be applied to
The gas constant is calculated from the ideal gas equation, R = the ratio uR2(R)/uc(R)2. The null hypothesis is uc(R)2 = uR2; it
pH2V/nH2T. The partial pressure of the hydrogen gas, pH2, is is sometimes accepted and sometimes rejected at the 95%
obtained by correcting patm for the hydrostatic pressure and confidence level, owing to the sensitivity of the ratio to small
subtracting pH2O, the vapor pressure of water at tsoln. The changes in the widths of the uncertainties of the measured
thermodynamic temperature of the hydrogen gas, T, is obtained quantities (particularly the mass of Mg and the volume of the
from t. Several assumptions are inherent in this calculation. gas in the buret). That uR2/uc(R)2 for the pooled data is nearly
They are examined in the Supporting Information and may be always greater than one not only draws attention to the spread
considered during the laboratory period, in the postlaboratory in the individual R values, which is not unexpected from a
discussion, or in the laboratory report. group of students with varying skill levels, but also suggests that
We spread the complete experiment over two sessions in a the estimates of the uncertainties in the individual measure-
course in which students are enrolled in the lecture and ments, specifically V, might be too small. In contrast with
laboratory concurrently. Students do the experiment in a 2-h titrations, where the buret is read twice, in this experiment the
laboratory period and complete the data analysis in a buret is read only once. In the former case, where students
subsequent 2-h lecture period. During the laboratory period, locate the meniscus is unimportant, so long as they locate it
students enter all their individual measurements in a spread- consistently, because the volume dispensed is found by
sheet. Students also enter the type (e.g., rectangular, triangular) subtraction. In this experiment, the differing judgments of
and width of the distribution for each measurement. The data students do matter and lead to a wider variability than one
set is posted on the class Web site, and students are asked to would expect based on the graduations marked on the buret.
compute values of R for the entire data set before the next class The ability to carry out a “what-if” analysis with the goal of
meeting. finding the range of uncertainties in V leading to acceptance of
A q-q plot,10 generated by the instructor, is also posted so the null hypothesis is a virtue of using a spreadsheet. However,
that students can visually compare the distribution of the set of the results of such an analysis must be checked against
R values to a normal distribution. With the exception of the reasonable limits derived from laboratory experience.
occasional outlier, student data sets follow a normal distribution The Supporting Information further describes the steps taken
reasonably closely. In six sets of pooled student data, with 11− in the analysis and illustrates them with a spreadsheet
28 individual trials, two sets had no outliers, three sets had one containing one complete set of student data, calculation of R,
outlier, and one set had two outliers. The familiar interpretation the q-q plot, application of Dixon’s test, calculation of the
of the mean and standard deviation are based on a normal standard and combined standard uncertainties, and application
distribution, so it is valuable for students to check the of the F test. The mean and standard uncertainty of the full data
expectation that their data are drawn from one. Moreover, set (N = 28) are R̅ = (0.081859 ± 0.000090) L atm mol−1 K−1.
approximate normality is a criterion for Dixon’s test and the F A q-q plot shows generally good agreement between the
B dx.doi.org/10.1021/ed3005374 | J. Chem. Educ. XXXX, XXX, XXX−XXX
Journal of Chemical Education


Laboratory Experiment

observed points and the reference line, with the point at the REFERENCES
lowest value of R being singled out for scrutiny as a possible (1) Taylor, B. N.; Kuyatt, C. E. Guidelines for Evaluating and
outlier. Expressing the Uncertainty of NIST Measurement Results; NIST TN
Dixon’s test13 using r22 supports discarding this point as an 1297; NIST: Gaithersburg, MD, 1994.
outlier. r22 is the ratio of the gap between the minimum or (2) Garland, C. W.; Nibler, J. W.; Shoemaker, D. P. Experiments in
maximum and its next nearest neighbor to a truncated range, Physical Chemistry, 8th ed.; McGraw-Hill: New York, 2009.
found by excluding the two largest or smallest values (the two (3) Halpern, A. M.; McBane, G. C. Experimental Physical Chemistry:
extreme values at the other end of the data set). The mean and A Laboratory Textbook, 3rd ed.; W. H. Freeman: New York, 2006.
(4) Sime, R. J. Physical Chemistry: Methods, Techniques and
standard uncertainty become R̅ = 0.081907 ± 0.000079 L atm Experiments; Saunders: Philadelphia, PA, 1990.
mol−1 K−1 (N = 27). The mean combined standard uncertainty (5) Lehman, T. A.; Harms, G. J. Chem. Educ. 1988, 65, 811−812.
is 0.000065 L atm mol−1 K−1. The mean is representative as the (6) Moss, D. B.; Cornely, K. J. Chem. Educ. 2001, 78, 1260−1262.
uc(R) are normally distributed, so it is used in the following (7) Donato, H., Jr.; Metz, C. J. Chem. Educ. 1988, 65, 867−868.
analysis; students use the values of uc(R) for each of their trials. (8) de Levie, R. J. Chem. Educ. 2000, 77, 534−535.
uR and uc(R ) are comparable, giving rise to relative uncertainties (9) Gardenier, G. H.; Gui, F.; Demas, J. N. J. Chem. Educ. 2011, 88,
in R of 0.00097 and 0.00079, respectively. The ratio F = uR2/ 916−920.
(10) Cleveland, W. S. Visualizing Data; Hobart Press: Summit, NJ,
(uc(R ))2 = 1.51 lies inside the range bounded by the lower and 1993.
upper critical values of F, which are 0.53 and 1.62 for α = 0.05, (11) Anderson, R. L. Practical Statistics for Analytical Chemists; Van
ν1 = 26 and ν2 = 5000; ν2 = 5000 approximates the convention Nostrand Reinhold: New York, 1987; pp 59−60.
ν2 = ∞.1,14 For this data set, the difference between uR and (12) NIST/SEMATECH e-Handbook of Statistical Methods: F-Test for
Equality of Two Variances. http://www.itl.nist.gov/div898/
uc(R ) is of neither practical nor statistical significance.
handbook/eda/section3/eda359.htm (accessed Mar 2013)
The similarity of these estimates of uncertainty confirms that (13) Rorabacher, D. B. Anal. Chem. 1991, 63, 139−146.
propagation of uncertainty is a viable alternative to statistical (14) Herberg, R. J. Anal. Chem. 1961, 33, 1308−11.
analysis. When determining an experimental quantity y, the
experimenter can expect uy and uc(y) to be comparable. If
obtaining a sufficient number of replicates is practical, uy is
chosen to express the uncertainty in y. It is the closest one can
come to the true standard deviation of the mean. If obtaining a
sufficient number of replicates is impractical, the experimenter
uses uc(y), confident that it is a reliable surrogate for uy.

■ SUMMARY
Determining the gas constant R by a classical method using
only standard equipment and measurement techniques affords
students the opportunity to focus on critical evaluation of the
data, unimpeded by the introduction of new, unfamiliar
chemical concepts. The simplicity of the experiment enables
creation of a sizable pooled data set that can be characterized by
conventional statistical analysis, with the standard deviation of
the mean providing one estimate of the uncertainty in the
measured value of R. Comparing this estimate to a second,
obtained via propagation of uncertainty, is the centerpiece of
the experiment, with the similarity of the two estimates
conferring credibility on the method of propagation of
uncertainty. The experiment lends itself to consideration of
more advanced statistical topics and experimental design,
allowing instructors to go beyond validating the method of
propagation of uncertainty at their discretion.

■ ASSOCIATED CONTENT
* Supporting Information
S

Spreadsheet for the class data; spreadsheet of the class results;


laboratory documentation including a handout for the students
and notes for the instructor. This material is available via the
Internet at http://pubs.acs.org.

■ AUTHOR INFORMATION
Corresponding Author
*E-mail: robert.olsen@stockton.edu.
Notes
The authors declare no competing financial interest.
C dx.doi.org/10.1021/ed3005374 | J. Chem. Educ. XXXX, XXX, XXX−XXX

You might also like