Home / Courses / GATE Physics / Practice / Measurement, errors and fitting data

Measurement, errors and fitting data

Experimental methods · 80 questions and templates for GATE Physics, in order from warm-up to top rank. Check each answer, then read the full solution.

Loading questions… (they need JavaScript; nothing you type is sent anywhere)

Written questions with solutions

The same questions as a list, so you can print them or read without the practice box. Templates (fresh numbers) appear only in the practice box above.

  1. MCQWarm-up

    Ten readings of a known 5.00 V reference give values from 5.31 V to 5.33 V. The voltmeter is
    1. accurate but not precise
    2. precise but not accurate
    3. both accurate and precise
    4. neither accurate nor precise
    Answer and solution

    Answer: B

    Precision is the spread of repeated readings: 0.02 V here, very small. Accuracy is closeness to the true value: the readings sit about 0.32 V too high. Small spread with a large offset means a systematic error (a calibration or zero error): precise but not accurate.

    Why people get this wrong: A small spread says nothing about the offset; averaging more readings will not remove it.

  2. MCQWarm-up

    Which of these reduces the random error of a result but leaves a systematic error untouched?
    1. Calibrating the instrument against a standard
    2. Correcting for the zero error of the scale
    3. Averaging a larger number of repeated readings
    4. Measuring the same quantity by a different method
    Answer and solution

    Answer: C

    The standard error of a mean falls as σ/N, so more readings reduce random scatter. A systematic error shifts every reading the same way, so it survives any amount of averaging; calibration, zero correction and an independent method are the ways to find and remove it.

    Why people get this wrong: Averaging never removes a bias.

  3. MCQWarm-up

    The sample standard deviation of N readings is computed with N−1 rather than N in the denominator because
    1. one degree of freedom is used up in estimating the mean from the same data
    2. it makes the standard deviation smaller
    3. the first reading is always discarded
    4. it converts the standard deviation into the standard error
    Answer and solution

    Answer: A

    Deviations are taken from the sample mean x‾, which is itself fitted to the data, so ∑(xi−x‾)2 is on average (N−1)σ2, not Nσ2. Dividing by N−1 gives an unbiased estimate of σ2. The same rule gives N−p degrees of freedom after fitting p parameters.

    Why people get this wrong: Dividing by N−1 makes the estimate slightly larger, not smaller.

  4. NATWarm-up

    Five readings of a length are 9.8, 10.1, 10.0, 10.3 and 9.8 cm. The standard error of their mean (using the sample standard deviation) is ______ cm (to three decimal places).
    Answer and solution

    Answer: 0.095 cm (±0.002)

    Mean =50.0/5=10.0 cm. Deviations: −0.2,0.1,0,0.3,−0.2; squares sum to 0.18. Sample standard deviation s=0.18/4=0.212 cm. Standard error s/5=0.212/2.236=0.095 cm.

    Why people get this wrong: The standard deviation (0.21 cm) describes one reading; the standard error describes the mean.

  5. MCQWarm-up

    For Gaussian errors, the fraction of readings expected within ±2σ of the mean is about
    1. 68.3%
    2. 90.0%
    3. 99.7%
    4. 95.4%
    Answer and solution

    Answer: D

    The probability within ±kσ is erf⁡(k/2): 68.3% for k=1, 95.4% for k=2, 99.7% for k=3.

    Why people get this wrong: 68% is one sigma; 99.7% is three.

  6. MCQWarm-up

    A digital meter shows whole millivolts. If the true value is equally likely to lie anywhere in the 1 mV interval the display rounds to, the standard uncertainty from this rounding alone is
    1. 0.5 mV
    2. 0.29 mV
    3. 1 mV
    4. 0.14 mV
    Answer and solution

    Answer: B

    A uniform distribution of full width Δ has standard deviation Δ/12. With Δ=1 mV: 1/3.464=0.29 mV. (Equivalently, half-width a=0.5 mV gives a/3.)

    Why people get this wrong: The half-width 0.5 mV is the maximum error, not the standard deviation.

  7. MCQWarm-up

    How many significant figures does the reading 0.004050 m have?
    1. 3
    2. 4
    3. 6
    4. 7
    Answer and solution

    Answer: B

    Leading zeros only place the decimal point, so they are not significant. The digits 4, 0, 5 and the final 0 are: the zero between 4 and 5 is enclosed, and a trailing zero after the decimal point records that the reading was made to that place. So 4 significant figures (4.050×10−3 m).

    Why people get this wrong: Dropping the final zero treats a measured digit as decoration.

  8. NATStandard

    A quantity f=x2y is formed from independent measurements with relative uncertainties 1% in x and 2% in y. The relative uncertainty of f is ______ % (to two decimal places).
    Answer and solution

    Answer: 2.83 % (±0.02)

    For products of powers, relative uncertainties add in quadrature, each multiplied by its power: σf/f=(2×1%)2+(1×2%)2=8%=2.83%.

    Why people get this wrong: Adding linearly (4%) is the worst case, not the standard uncertainty for independent errors.

  9. MCQStandard

    Two masses are A=50.0±0.3 g and B=48.0±0.4 g (independent). The relative uncertainty of the difference A−B is
    1. 35%
    2. 1.4%
    3. 25%
    4. 0.7%
    Answer and solution

    Answer: C

    A−B=2.0 g with σ=0.32+0.42=0.5 g, so the relative uncertainty is 0.5/2.0=25%. Subtracting nearly equal numbers keeps the absolute error but destroys the relative precision.

    Why people get this wrong: 35% comes from adding 0.3 and 0.4 linearly; 1% or less comes from using the relative errors of A and B.

  10. MCQStandard

    For data (xi,yi±σi) with negligible errors in x, the least-squares fit of a model y(x) chooses the parameters that minimise
    1. ∑i|yi−y(xi)|
    2. ∑i(yi−y(xi))/σi
    3. ∑i(xi−x‾)2
    4. ∑i[(yi−y(xi))/σi]2
    Answer and solution

    Answer: D

    For Gaussian errors the likelihood is ∏e−(yi−y(xi))2/2σi2, which is largest when χ2=∑[(yi−y(xi))/σi]2 is smallest. Points with small σi get more weight.

    Why people get this wrong: The plain sum of residuals can be zero for a terrible fit; signs cancel.

  11. NATStandard

    A least-squares straight line y=a+bx is fitted to (1,2.1), (2,3.9), (3,6.2), (4,7.8). The intercept a is ______ (to two decimal places).
    Answer and solution

    Answer: 0.15 (±0.01)

    N=4, ∑x=10, ∑y=20.0, ∑xy=59.7, ∑x2=30. Δ=N∑x2−(∑x)2=20. b=(4×59.7−10×20)/20=1.94. The line passes through (x‾,y‾)=(2.5,5.0), so a=5.0−1.94×2.5=0.15.

    Why people get this wrong: Every least-squares line with an intercept passes through the centroid (x‾,y‾); use it to get a quickly.

  12. MCQStandard

    The current in a discharging circuit is expected to follow I=I0e−t/τ. Which plot is a straight line, and what does its slope give?
    1. ln⁡I against t; slope −1/τ
    2. I against ln⁡t; slope −τ
    3. ln⁡I against ln⁡t; slope −1/τ
    4. 1/I against t; slope τ
    Answer and solution

    Answer: A

    Taking logs, ln⁡I=ln⁡I0−t/τ: linear in t with slope −1/τ and intercept ln⁡I0.

    Why people get this wrong: A log-log plot linearises power laws, not exponentials.

  13. MCQStandard

    Data are suspected to follow y=Cxn with unknown C and n. A straight-line fit of which quantities gives n directly as the slope?
    1. y against x2
    2. ln⁡y against x
    3. ln⁡y against ln⁡x
    4. 1/y against 1/x
    Answer and solution

    Answer: C

    ln⁡y=ln⁡C+nln⁡x: slope n, intercept ln⁡C. Any base of logarithm works for the slope.

    Why people get this wrong: ln⁡y against x is for exponentials.

  14. MCQStandard

    A straight line (two parameters) fitted to 12 points with known errors gives χ2=30. The reduced chi-square is
    1. 2.5
    2. 2.7
    3. 3.0
    4. 15
    Answer and solution

    Answer: C

    Degrees of freedom ν=N−p=12−2=10, so χ2/ν=3.0: the scatter is about 3 times larger than the stated errors explain.

    Why people get this wrong: Dividing by N (2.5) or N−1 (2.7) forgets that two parameters were fitted.

  15. MSQStandard

    Which of these are systematic errors?
    1. A screw gauge whose jaws read 0.02 mm when closed
    2. Reading a needle always from slightly to the left of the scale
    3. The last digit of a voltmeter flickering between two values
    4. A steel tape calibrated at 20 °C used outdoors at 40 °C
    Answer and solution

    Answer: A, B, D

    A zero error, a consistent parallax and thermal expansion of the tape all push every reading in the same direction: systematic. A flickering last digit scatters readings both ways: random.

    Why people get this wrong: Parallax is systematic when it is always from the same side; random only if the eye position varies.

  16. MCQStandard

    A calculation gives g=9.8164 m s−2 with a standard uncertainty of 0.0372 m s−2. The correct way to report it is
    1. 9.8164±0.0372 m s−2
    2. 9.82±0.04 m s−2
    3. 9.8±0.0372 m s−2
    4. 9.816±0.04 m s−2
    Answer and solution

    Answer: B

    Round the uncertainty to one (or two) significant figures, here 0.04, then round the value to the same decimal place: 9.82. The digits beyond are not known.

    Why people get this wrong: The value and its uncertainty must end at the same decimal place.

  17. MCQStandard

    A counter records 900 counts in 300 s (background negligible). The count rate is
    1. 3.0±0.1 s−1
    2. 3.0±1.7 s−1
    3. 3.0±0.03 s−1
    4. 3.0±30 s−1
    Answer and solution

    Answer: A

    Counts are Poisson: σN=900=30. The time is exact, so R=(900±30)/300=3.0±0.1 s−1.

    Why people get this wrong: Taking R=1.7 is wrong: the square root rule applies to the number of counts, not to a rate.

  18. MCQStandard

    A detector registers 80 of 100 particles sent through it. Its efficiency, with its standard uncertainty, is
    1. 0.80±0.04
    2. 0.80±0.09
    3. 0.80±0.08
    4. 0.80±0.004
    Answer and solution

    Answer: A

    The number detected out of a fixed n is binomial, with variance np(1−p). So σp=p(1−p)/n=0.8×0.2/100=0.04. Using Poisson errors (80/100=0.09) overstates it, and would wrongly give a non-zero error even at p=1.

    Why people get this wrong: N errors belong to counts with no fixed total; an efficiency is a fraction of a fixed number of trials.

  19. NATStandard

    A resistance is found from V=12.0±0.1 V and I=2.00±0.02 A (independent errors) as R=V/I. The standard uncertainty of R is ______ Ω (to three decimal places).
    Answer and solution

    Answer: 0.078 (±0.001)

    R=6.00 Ω. For a quotient the relative errors add in quadrature: (0.1/12.0)2+(0.02/2.00)2=6.94×10−5+1.00×10−4=0.0130. So σR=0.0130×6.00=0.078 Ω.

    Why people get this wrong: Adding the relative errors linearly (1.83%) gives 0.11 Ω, the worst case, not the standard uncertainty.

  20. MCQStandard

    A quantity is measured as x=200±10. The value of ln⁡x with its standard uncertainty is
    1. 5.30±2.30
    2. 5.30±10
    3. 5.30±0.69
    4. 5.30±0.05
    Answer and solution

    Answer: D

    ln⁡200=5.30 and d(ln⁡x)/dx=1/x, so σln⁡x=σx/x=10/200=0.05. The absolute error of a logarithm equals the relative error of its argument, which is why log plots show equal relative errors as equal bars.

    Why people get this wrong: Taking the logarithm of the error (ln⁡10=2.30) is not error propagation.

  21. NATExam level

    A pendulum has length L=1.000±0.002 m. Fifty oscillations take 100.0±0.2 s. Using g=4π2L/T2, the uncertainty in g is ______ m s−2 (to three decimal places).
    Answer and solution

    Answer: 0.044 m s⁻² (±0.002)

    T=2.000±0.004 s (timing 50 swings divides the error by 50). g=4π2(1.000)/4=9.870 m s−2. σg/g=(0.002)2+(2×0.002)2=4.47×10−3, so σg=0.044 m s−2.

    Why people get this wrong: The period enters squared, so its relative error counts twice.

  22. MCQExam level

    Two lengths L1 and L2 are measured with the same tape, which has an unknown scale error of 0.5% (every reading too long by the same factor). The scale error contributes to the ratio L1/L2 an uncertainty of
    1. 0.71%
    2. 0.5%
    3. 1.0%
    4. zero
    Answer and solution

    Answer: D

    Both readings carry the same factor (1+ϵ), so L1(1+ϵ)/[L2(1+ϵ)]=L1/L2: the error cancels exactly. Fully correlated errors must be propagated with their covariance, not in quadrature.

    Why people get this wrong: 0.71% treats the two scale errors as independent, which they are not.

  23. NATExam level

    Three labs report 4.2±0.2, 4.6±0.4 and 4.0±0.1 (same units). Their weighted mean is ______ (to two decimal places).
    Answer and solution

    Answer: 4.07 (±0.01)

    Weights w=1/σ2: 25, 6.25, 100 (sum 131.25). x‾=(4.2×25+4.6×6.25+4.0×100)/131.25=533.75/131.25=4.067.

    Why people get this wrong: The plain average 4.27 ignores that the third lab is four times more precise than the second.

  24. MCQExam level

    For the three results 4.2±0.2, 4.6±0.4 and 4.0±0.1, the uncertainty of the weighted mean is
    1. 0.10
    2. 0.23
    3. 0.087
    4. 0.058
    Answer and solution

    Answer: C

    σx‾=(∑1/σi2)−1/2=131.25−1/2=0.087. It is smaller than the best single error (0.1), as it must be.

    Why people get this wrong: The combined result can never be less precise than its best input.

  25. NATExam level

    A straight line is fitted to five points at x=1,2,3,4,5, each y having the same standard error 0.2. The standard error of the fitted slope is ______ (to three decimal places).
    Answer and solution

    Answer: 0.063 (±0.001)

    σb=σN/Δ with Δ=N∑x2−(∑x)2=5×55−152=50. σb=0.25/50=0.2×0.316=0.063. Equivalently σ/∑(xi−x‾)2=0.2/10.

    Why people get this wrong: The slope error does not depend on the y values, only on σ and the spread of the x values.

  26. MCQExam level

    After a straight-line fit, the residuals are positive at both ends of the range and negative in the middle, although χ2/ν is close to 2. The most likely conclusion is
    1. the error bars are too small by a factor of 2
    2. the data have curvature that a straight line cannot describe
    3. there is one outlier
    4. the fit is good and the pattern is chance
    Answer and solution

    Answer: B

    Random errors give residuals that change sign at random. A smooth U-shaped pattern is systematic: the model is missing a term (for instance cx2). Fit a better model and look again before blaming the error bars.

    Why people get this wrong: Inflating the error bars would lower χ2 but leave the pattern, and the physics, wrong.

  27. MCQExam level

    A fit with 20 degrees of freedom gives χ2=4. The most reasonable interpretation is
    1. the model is wrong
    2. the fit is excellent and the errors are correct
    3. the data should be refitted with more parameters
    4. the stated errors are too large (or the points are not independent)
    Answer and solution

    Answer: D

    χ2/ν=0.2, while a good fit with correct errors gives about 1±2/ν=1±0.32. Such a small value means the scatter is much smaller than the errors claim: errors overestimated or correlated (for example, smoothed data).

    Why people get this wrong: A very small χ2 is as suspicious as a very large one.

  28. NATExam level

    A theory predicts 10.0; an experiment gives 10.6±0.25 (Gaussian). The probability that chance alone gives a deviation at least this large in either direction is ______ % (to two decimal places). Use Φ(2.4)=0.9918 for the standard normal cumulative distribution.
    Answer and solution

    Answer: 1.64 % (±0.05)

    z=(10.6−10.0)/0.25=2.4. Two-sided probability =2[1−Φ(2.4)]=2×0.0082=0.0164=1.64%. The disagreement is significant at the 5% level but not at the 1% level.

    Why people get this wrong: One-sided (0.82%) is the answer to a different question.

  29. MSQExam level

    For an unweighted least-squares straight line y=a+bx through N points, which statements are true?
    1. The line passes through (x‾,y‾)
    2. The residuals add up to zero
    3. The fitted a and b are uncorrelated whatever the x values
    4. ∑(residual)2/(N−2) estimates the variance of a single y
    Answer and solution

    Answer: A, B, D

    Setting ∂S/∂a=0 gives ∑(yi−a−bxi)=0: residuals sum to zero and the line passes through the centroid. With two fitted parameters, ∑r2/(N−2) is the unbiased variance estimate. cov(a,b)=−x‾σb2, which is zero only when x‾=0.

    Why people get this wrong: Shift the origin to x‾ if you want uncorrelated parameters.

  30. MCQExam level

    Decay data are fitted as ln⁡N against t. A point with N=625 counts (Poisson) has an uncertainty in ln⁡N of
    1. 25
    2. 0.04
    3. 0.0016
    4. 3.2
    Answer and solution

    Answer: B

    σln⁡N=σN/N=N/N=1/N=1/25=0.04. Points with fewer counts have larger errors in ln⁡N, so a correct fit weights each point by N.

    Why people get this wrong: Taking logs changes the error bars: equal weights on a log plot are wrong for counting data.

← All topics · Timed mock test →