Академический Документы
Профессиональный Документы
Культура Документы
Significant Figures
Rules
• addition and subtraction
—identify common decimal point
• multiplication and division
—limited to the least number of
significant figures
• logs and antilogs
—limited to number of digits in
mantissa
1
Significant Figures Significant Figures
Sig Figs in Graphs • Data in the lab deals with numbers
100000
• Rules have been developed to express the error
associated with each number and the goodness of
80000
each number
60000
40000
Example: pan balance----error of 0.01 g
20000
0
0 .5 1 2 .5 5
• a 100.0000 g item that is weighed can only be
S h e a r R a t e ( rp m )
reported as 100.00 g
• digits beyond 100.00 are not significant
• Significance of data is described through
• data only has 5 significant figures
use of error bars
2
Statistical Data Treatment Statistical Data Treatment
Reading Ch7
Confidence Level (CL) z: probability factor
--Define an interval around an experimentally (the deviation from the mean in units of σ)
CL z zσ
determined mean that probably contains the CI = µ ±
population mean µ (probability that the true mean 50 0.67 N
lies in a certain interval) 68 1.00
80 1.29 •N is a very large number
•A common calculation to report variability of data
90 1.65 •Z comes from the infinity
•A quick way of identifying outlying values
95 1.96 row of the t value
Confidence Interval (CI):The range of values within 99.99995 5.00
which the population mean µ is expected to lie at
certain CL.
zσ For small data sets (much more common)
For large data sets CI = µ ± We can rely on the t test.
N
3
t Values
x−µ x−µ
t= vs z=
s σ
•Like z, t depends on the desired CL
•t values account for error introduced based on
sample size, degree of freedom and potential
sample skew. ts zσ
CI = x ± CI = µ ±
N N
s: standard deviation of the mean
•when N→∞, t→z. (Table 7-3)
•At the same desired CL, N↑, t↓
4
Comparing Results Comparing two experimental means
Two sets of numbers
More easier way to compare
--- from different samples
x − xt N --- from different assays of the same sample
tcalc. = •They are actually different.
s
•If the means are identical, it is more likely an
If tcalc. > tcrit., then the difference is significant accident than anything else.
To test if the two means actually differ, first
Again, depending on the desired CL you must calculate the mean and standard
deviation for each set.
x1 − x = ± ts (N + N ) N N
x1 − x 2 = 5 mg / L
2 p 1 2 1 2
4. Pick t based on desired confidence level and df Are the two sets of data significantly different at 95%
(df = nA + nB –2), If x 1 − x 2 > V crit . CL?
Then the means are different.
5
Steps to follow The F test
The F test can be used to tell if the precision of two
1. s = 2 × 4 + 1.5 × 5 = 1.74
2 2
p
4+5 sets of measurement are different.
2. x1 − x 2 5 •Has the measurement precision changed?
t calc. = N 1 N 2 /( N 1 + N 2 ) = 5 × 6 /(5 + 6) = 4.74
sp 1.74
•Has the method been altered?
3. Use 95% confidence limit (t = 2.26) •Were there any significant changes due to the
lab or analyst?
tcalc. > tcrit .
Sl2arg er
So the means are different at the 95% confidential level. Calculation of F F = 2
S smaller
F ≥ 1 and depends on the confidence level and df.
NOTE: if t calc . < t crit . , it does not mean that two
You can look up the Fc value for the desired levels
means are equal,it means no significant difference
(Table 7-4).
The F test
Example
A-- s = 0.15 ppm, df =12
B-- s = 0.12 ppm, df =20
F = 0.152/0.122 = 1.56
Fc is 2.28 at 95% confidence.
F < Fc, one cannot say method A is better than
method B in terms of the precision
6
Rejection of data Q test
Sometimes we know that a data point looks Assumes
bad(outlier).
• x and Sx are unknown.
Outliers
• Data is normally distributed.
Q test
Example: The analysis of a calcite sample yielded
CaO percentages of 55.95, 56.04, 56.08, 56.23, and
56.00. The values of 56.23 appears anomalous, should
it be rejected at the 95% confidence level?
7
Standardization and Calibration Method of Least Squares
•Instruments and techniques produce signals, not •Assuming there is a linear relationship between the
concentrations. measured variable (y) and the analyte conc. (x)
•Analyst must correlate the signal level to a y = mx + b
corresponding concentration---Calibration.
•The working curve is produced through regression • Assuming that any deviation of individual points
analysis to find best fit from the straight line is from the error of
measurement
1. Comparison with Standards In other words, no error from x value (knowing the
2. External Standard Calibration exact conc. of standard samples)
Regression Analysis
Residual = yi-(mxi+b) For the linear equation:
y = mx + b
Useful quantities:
Slope:
Intercept:
8
Regression Analysis Example
9
Quality assurance and control chart Quality Control Chart
--Define the upper control limit(UCL) and the lower
Two types of quality assessment measurements
control limit(LCL)
*Evaluation of the accuracy and precision of the 3σ 3σ
UCL = µ + LCL = µ −
methods of measurement N N
*Evaluation of product quality (e.g., F in toothpaste)
µ----population mean for the mass measurement
How to assess the quality σ----population standard deviation
---usage of quality control chart
N----the number of replicates
Example: control chart for a analytical balance Example: Control chart for monitoring the conc.
of benzoyl peroxide in a commerical product
10