GELE Adjustment Computations (Least Squares) — Error Propagation, Variance-Covariance and Error EllipsesMisconception Buster
Mistake patterns in Error Propagation, Variance-Covariance and Error Ellipses — the trap questions GELE sets and the wrong assumptions reviewers make. This page walks through each misconception, why it is wrong, and how Professional Regulation Commission (PRC) — Board of Geodetic Engineering turns it into a tempting but incorrect answer choice.
Exam context
On the GELE 2026, the Adjustment Computations (Least Squares) subtest carries a "Core" weight in Professional Regulation Commission (PRC) — Board of Geodetic Engineering's pattern. Error Propagation, Variance-Covariance and Error Ellipses lands at position 5th out of 5 in the standard review order. Target score is 70% weighted average, no sub-test below 50%, and roughly a meaningful share of items come from Adjustment Computations (Least Squares) on a typical GELE paper.
Error Propagation, Variance-Covariance and Error Ellipses - Misconception Buster
In the PRC Geodetic Engineer Licensure Examination, the topic of Error Propagation, Variance-Covariance, and Error Ellipses consistently separates passing candidates from failing ones. These concepts are mathematically precise — a single wrong formula, a misapplied sign rule, or a confused interpretation of the error ellipse can cascade into completely wrong answers across multiple items. What makes this chapter especially dangerous is that many misconceptions feel intuitively correct. Students who learned only the surface-level rules without understanding the underlying statistical theory are most at risk. This guide targets exactly those wrong beliefs: it names them, explains why they feel right, proves why they are wrong, and gives you a trap question — the kind of question that appears on actual board exams and is specifically designed to catch students who hold these misconceptions. Study this guide not just to review formulas, but to rewire your thinking before exam day.
Summary
The ten most dangerous misconceptions in Error Propagation, Variance-Covariance, and Error Ellipses share a common root cause: students apply 1D intuitions and linear thinking to inherently quadratic, matrix-based, and 2D statistical concepts. The five most exam-critical takeaways are: (1) Independent errors ALWAYS combine in quadrature — never linearly. (2) Distinguish σ√n (sum of n segments) from σ/√n (mean of n readings) — they move in opposite directions. (3) The standard error ellipse is ~39%, not 68% or 95% — scale by 2.45 for the 95% confidence ellipse using the 2D chi-squared factor, not the 1D z-score of 1.96. (4) The diagonal of the covariance matrix contains variances (m²), not standard deviations (m) — always take the square root for σ. (5) Larger weight means higher precision (smaller σ), not lower precision — the inverse variance relationship is absolute. Beyond these, remember that: the matrix propagation formula Σ_y = JΣ_xJᵀ is required whenever inputs are correlated; the a posteriori variance factor σ̂₀² ≠ 1 in general and must multiply (AᵀPA)⁻¹; the semi-axes of the error ellipse are √eigenvalues, not eigenvalues themselves; a circular ellipse indicates good geometric conditioning, not necessarily high accuracy; and the ellipse orientation depends on eigenvectors, not coordinate grid alignment. Master these distinctions and the most common exam traps in this chapter will become immediately recognizable.
Misconceptions
When combining independent errors, you simply add the individual standard deviations: σ_total = σ₁ + σ₂.
Tags
- critical_error
- formula_confusion
- most_common_mistake
- quadrature_rule
Topic
Law of Propagation of Variances — Sum/Difference Rule
Severity
critical
Exam Impact
This misconception directly causes wrong numerical answers. For two measurements each with σ = 0.03 m, linear addition gives σ_total = 0.06 m while the correct answer is √(0.03² + 0.03²) = 0.0424 m. That is a 41% overestimate — enough to select a completely wrong multiple-choice option.
The Reality
Independent errors NEVER add linearly — they add in quadrature (root-sum-of-squares). This comes directly from the Law of Propagation of Variances: σ_y² = (∂f/∂x₁)²σ₁² + (∂f/∂x₂)²σ₂². For y = x₁ + x₂, both partial derivatives equal 1, giving σ_y² = σ₁² + σ₂², so σ_y = √(σ₁² + σ₂²). Linear addition would be correct only if errors were perfectly correlated (ρ = +1), which is not the case for independent measurements. The quadrature rule means combined uncertainty is ALWAYS less than the linear sum — a critically important distinction.
Trap Question
Question
Two traverse segments are measured independently. Segment AB has σ = 0.04 m and segment BC has σ = 0.03 m. What is the standard deviation of the total distance AC?
Explanation
The two segments are measured independently, so their errors are uncorrelated. The Law of Propagation of Variances requires summing the variances (squared standard deviations), then taking the square root. Note that in this special case (a 3-4-5 Pythagorean triple in the σ domain), the answer is exactly 0.05 m — a result that board exams frequently use precisely because linear addition (0.07 m) is so tempting and so wrong.
Wrong Answer
σ_AC = 0.04 + 0.03 = 0.07 m
Correct Answer
σ_AC = √(0.04² + 0.03²) = √(0.0016 + 0.0009) = √0.0025 = 0.05 m
Misconception Id
M1
Correct Vs Incorrect
Correct Approach
σ_total = √(σ₁² + σ₂²) = √(0.03² + 0.03²) = √(0.0009 + 0.0009) = √0.0018 = 0.0424 m ← CORRECT. Errors are independent, so variances (not standard deviations) are summed, then square-rooted.
Incorrect Approach
σ_total = σ₁ + σ₂ = 0.03 + 0.03 = 0.06 m ← WRONG. This assumes perfect positive correlation between the two errors, which violates the independence assumption.
Why Students Believe It
Addition feels natural because we are literally combining two measurements. Students memorize 'errors add' without internalizing what is actually being added. The word 'sum' in everyday language means direct addition, so applying that instinct to uncertainties seems obvious and correct.
The standard error of the mean and the standard error of the sum both use the formula σ√n.
Tags
- critical_error
- formula_confusion
- sigma_sqrt_n
- mean_vs_sum
Topic
Error Propagation — Sum vs. Mean
Severity
critical
Exam Impact
Exam problems often describe a scenario that could be interpreted as either summing or averaging. A student with this misconception will apply σ√n when σ/√n is needed, or vice versa, producing an answer that is wrong by a factor of n — a very large numerical error that eliminates the correct choice entirely.
The Reality
These two formulas are exact opposites in effect. For the SUM of n equal, independent observations: σ_sum = σ√n (uncertainty GROWS with n — more segments means more accumulated error). For the MEAN of n equal, independent observations: σ_mean = σ/√n (uncertainty SHRINKS with n — more repetitions improve precision). The physical intuition: taping a distance in 6 segments accumulates 6 independent errors (sum), while averaging 6 independent readings of the same quantity averages out random errors (mean). Confusing these two is catastrophically wrong because they move in opposite directions.
Trap Question
Question
A levelling line is run in 9 equal sections. Each section has a standard error of 3 mm. What is the standard error of the total levelled height difference across all 9 sections?
Explanation
Each section is a separate, independent measurement being ADDED to form the total. This is a SUM of 9 independent errors, so σ_sum = σ√n = 3√9 = 9 mm. The formula σ/√n applies only when you take n readings of THE SAME quantity and average them. In levelling, each section measures a DIFFERENT height difference — they are summed, not averaged. This distinction is tested repeatedly in Philippine board exams on geodetic levelling.
Wrong Answer
σ = 3/√9 = 1 mm (applying the mean formula instead of the sum formula)
Correct Answer
σ_total = 3√9 = 3 × 3 = 9 mm
Misconception Id
M2
Correct Vs Incorrect
Correct Approach
σ_mean = σ/√n = 0.02/√4 = 0.02/2 = 0.01 m. Averaging 4 readings halves the standard error. In contrast, if the distance is taped in 4 equal segments (sum), σ_sum = 0.02√4 = 0.04 m. The key distinction: Are we COMBINING (summing) independent segments, or AVERAGING repeated observations of the same quantity?
Incorrect Approach
A distance is measured 4 times, each with σ = 0.02 m. The student writes σ_mean = 0.02√4 = 0.04 m. WRONG — this is the formula for the sum, not the mean. The student has made the uncertainty LARGER by averaging, which contradicts the entire purpose of repeated measurement.
Why Students Believe It
Students often memorize 'something involves √n and σ' without distinguishing WHICH statistic is being computed. Both the sum and the mean involve n and σ, creating a memory-blending confusion. Under exam pressure, the distinction collapses.
The standard error ellipse represents the 95% confidence region for a 2D position.
Tags
- critical_error
- conceptual_gap
- confidence_level
- 2D_statistics
- chi_squared
Topic
Error Ellipse — Confidence Levels and Scaling
Severity
critical
Exam Impact
Board exam questions may ask for the 95% confidence ellipse semi-axis given the standard ellipse semi-axis, or may ask what probability is enclosed by the standard ellipse. A student with this misconception will report the standard ellipse as 95% (losing direct marks) or scale by the wrong factor (1.96 instead of 2.45, a 1D normal quantile instead of the 2D chi-squared root).
The Reality
The standard error ellipse (semi-axes = eigenvalues of the 2×2 covariance matrix) corresponds to approximately 39% probability — NOT 68% and certainly not 95%. This is a fundamental result of 2D chi-squared statistics: the 1σ ellipse in 2D encloses only about 39% of the probability mass. To reach 95% confidence in 2D, you must scale the ellipse by a factor of approximately 2.45 (since the chi-squared quantile for 2 degrees of freedom at 95% gives χ²₀.₉₅,₂ = 5.991, and √5.991 ≈ 2.45). For 99%, the scale factor is √9.210 ≈ 3.03.
Trap Question
Question
A least-squares adjustment yields a standard error ellipse with semi-major axis a = 0.06 m and semi-minor axis b = 0.04 m for an adjusted control point. What are the semi-axes of the 95% confidence ellipse?
Explanation
In 2D, the joint probability distribution follows a chi-squared distribution with 2 degrees of freedom. The 95th percentile of χ²(2) is 5.991, giving a scale factor of √5.991 ≈ 2.45 — not the 1D factor of 1.96. Using 1.96 (the 1D z-score for 95%) underestimates the confidence region. This distinction between 1D and 2D confidence intervals is a classic board exam discriminator.
Wrong Answer
a₉₅ = 1.96 × 0.06 = 0.118 m; b₉₅ = 1.96 × 0.04 = 0.078 m (using the 1D normal 95% factor)
Correct Answer
Scale factor = √5.991 ≈ 2.45; a₉₅ = 2.45 × 0.06 = 0.147 m; b₉₅ = 2.45 × 0.04 = 0.098 m
Misconception Id
M3
Correct Vs Incorrect
Correct Approach
The standard ellipse (semi-axes a, b) encloses ~39% probability. To obtain the 95% confidence ellipse, multiply semi-axes by √χ²₀.₉₅,₂ = √5.991 ≈ 2.45. So a₉₅ = 2.45 × 0.05 = 0.1225 m and b₉₅ = 2.45 × 0.03 = 0.0735 m. For a 99% ellipse, use √9.210 ≈ 3.03.
Incorrect Approach
Student sees semi-axes a = 0.05 m, b = 0.03 m and reports these as the 95% confidence ellipse boundaries. Or, student scales by 1.96 (the 1D 95% z-score): a₉₅ = 1.96 × 0.05 = 0.098 m. Both are WRONG — the first because 1σ ellipse is only 39%, the second because 1.96 is a 1D normal quantile, not the 2D chi-squared scaling factor.
Why Students Believe It
In 1D statistics, the standard deviation (±1σ) is associated with approximately 68% confidence, and the 95% level is the one most used in practice. Students assume the same 68% rule carries over to 2D, and since 68% rounds to 'about two-thirds' (close enough to seem reasonable), the error ellipse is loosely associated with a high-confidence boundary.
In the propagation formula σ_y² = Σ(∂f/∂xᵢ)²σᵢ², you can ignore the partial derivative if it equals a simple constant like 1 or 2.
Tags
- major_error
- formula_confusion
- partial_derivative
- evaluation_error
Topic
Law of Propagation of Variances — Partial Derivatives
Severity
major
Exam Impact
When computing the propagated uncertainty of a derived quantity (area, volume, angle from components), forgetting to square the partial derivative or applying it linearly instead of quadratically will produce a wrong numerical answer. This affects problems involving areas from measured dimensions, computed angles, and coordinate transformations.
The Reality
The partial derivative is ALWAYS squared in the propagation formula, and its value always matters. For ∂f/∂xᵢ = 1, the term becomes (1)²σᵢ² = σᵢ², which is correct and must be included. For ∂f/∂xᵢ = 2 (e.g., for A = s² at s = 50 m, ∂A/∂s = 2s = 100), the term is (100)²σ_s², which amplifies the variance by 10,000. This is the mechanism by which a small measurement error can become a large area or volume error. Ignoring or under-weighting partial derivatives is a systematic error in propagation calculations.
Trap Question
Question
The area of a square lot is computed as A = s². The side s = 50 m is measured with σ_s = 0.02 m. What is the standard deviation of the computed area?
Explanation
The partial derivative ∂A/∂s = 2s must be evaluated at the actual measurement value s = 50 m, giving 100 — not just 2. The factor 2 alone is the symbolic derivative before substituting s. The correct propagation gives σ_A = 100 × 0.02 = 2.0 m². The wrong answer of 0.04 m² is 50 times too small, which would be a catastrophically wrong estimate of area uncertainty in cadastral surveys under PD 1529.
Wrong Answer
σ_A = 2 × 0.02 = 0.04 m² (applying the factor 2 linearly without squaring or evaluating the derivative at s = 50)
Correct Answer
∂A/∂s = 2s = 2(50) = 100; σ_A = |2s|σ_s = 100 × 0.02 = 2.0 m²
Misconception Id
M4
Correct Vs Incorrect
Correct Approach
∂A/∂L = W, ∂A/∂W = L. By propagation: σ_A² = W²σ_L² + L²σ_W². For L = 80 m (σ_L = 0.03 m) and W = 40 m (σ_W = 0.02 m): σ_A² = (40)²(0.03)² + (80)²(0.02)² = 1600(0.0009) + 6400(0.0004) = 1.44 + 2.56 = 4.00 m⁴. Therefore σ_A = √4.00 = 2.0 m².
Incorrect Approach
Area A = L × W. Student writes σ_A = W·σ_L + L·σ_W (linear sum of scaled uncertainties). This would be correct only if errors were perfectly correlated AND we were not squaring — it violates the propagation law on both counts for independent measurements.
Why Students Believe It
Students think that a partial derivative of 1 means 'nothing changes' and so it drops out of the calculation. For the derivative equal to 2, students might think it just doubles the uncertainty linearly. The act of squaring the derivative is often forgotten because students rush through the formula and drop the squared term.
The diagonal elements of the variance-covariance matrix Σ represent the standard deviations of the adjusted parameters, not the variances.
Tags
- major_error
- matrix_interpretation
- variance_vs_stddev
- covariance_matrix
Topic
Variance-Covariance Matrix — Diagonal Elements
Severity
major
Exam Impact
When a problem provides the covariance matrix and asks for the standard error of a coordinate, a student with this misconception will report √(σ²) correctly but may skip the square root and use the raw diagonal value — overstating the uncertainty by a factor of σ. In matrix-form propagation, using Σ elements as standard deviations will corrupt all subsequent calculations.
The Reality
The diagonal elements of the variance-covariance matrix Σ_x̂ = σ̂₀²(AᵀPA)⁻¹ are VARIANCES, not standard deviations. Specifically, Σᵢᵢ = σ²(x̂ᵢ). To obtain the standard deviation of an adjusted parameter, you must take the square root: σ(x̂ᵢ) = √Σᵢᵢ. The off-diagonal elements Σᵢⱼ are covariances, and their square roots (after proper normalization) give correlation coefficients. This distinction is fundamental: treating σ² as σ will give wrong propagated uncertainties and wrong confidence ellipse dimensions.
Trap Question
Question
After a least-squares adjustment of a control survey, the 2×2 variance-covariance submatrix for an adjusted point is: Σ = [[0.0025, 0.0010], [0.0010, 0.0016]] (all in m²). What is the standard error of the Northing coordinate?
Explanation
The element Σ₂₂ = 0.0016 is the VARIANCE of the Northing coordinate (units: m²). The standard deviation is its square root: σ_N = √0.0016 = 0.04 m. The off-diagonal element 0.0010 m² is the covariance between Easting and Northing, not directly a standard deviation. This distinction is essential when reporting coordinate accuracy for cadastral surveys (PD 1529) and NAMRIA control network submissions.
Wrong Answer
σ_N = 0.0016 m (reading the diagonal element directly as standard deviation)
Correct Answer
σ_N = √0.0016 = 0.04 m
Misconception Id
M5
Correct Vs Incorrect
Correct Approach
Diagonal elements are variances: σ²_x = 0.0004 m² and σ²_y = 0.0009 m². Standard deviations are: σ_x = √0.0004 = 0.02 m and σ_y = √0.0009 = 0.03 m. These are the physically meaningful measures of positional uncertainty reported for adjusted PPCS/UTM coordinates.
Incorrect Approach
The covariance matrix diagonal gives [0.0004, 0.0009] for a 2D point. Student reports σ_x = 0.0004 m and σ_y = 0.0009 m. These are variances, not standard deviations. The student has underreported precision by not taking the square root.
Why Students Believe It
Students are more accustomed to working with standard deviations (σ) than variances (σ²) in everyday computations. The covariance matrix is dense with notation, and under exam stress, students read diagonal values as σ and use them directly in subsequent calculations without taking the square root where needed.
A circular (or near-circular) error ellipse means the survey has high accuracy, while an elongated ellipse means the survey has poor accuracy overall.
Tags
- major_error
- conceptual_gap
- ellipse_interpretation
- geometric_conditioning
Topic
Error Ellipse — Shape and Geometric Conditioning
Severity
major
Exam Impact
Exam questions on error ellipse interpretation test whether students understand what geometric conditioning means versus absolute accuracy. A student with this misconception will misidentify which survey has better quality or misinterpret what the major axis direction signifies, losing conceptual marks.
The Reality
The shape of the error ellipse describes GEOMETRIC CONDITIONING, not absolute accuracy. A circular ellipse means the position uncertainty is the same in all directions — the network is well-conditioned geometrically (no preferred direction of weakness). An elongated ellipse means the position is well-determined in the direction of the minor axis but poorly determined in the direction of the major axis. This reflects the geometry of the observations (e.g., all measurements coming from one direction). A survey can have a circular ellipse that is LARGE (poor accuracy uniformly) or an elongated ellipse with a very small minor axis (excellent accuracy in one direction). You must look at BOTH shape AND size: the size of the semi-axes tells you about the magnitude of uncertainty; the shape tells you about its directional distribution.
Trap Question
Question
Point P has a standard error ellipse with semi-major axis a = 0.08 m and semi-minor axis b = 0.07 m. Point Q has a standard error ellipse with semi-major axis a = 0.20 m and semi-minor axis b = 0.02 m. Which point has better-conditioned network geometry?
Explanation
Geometric conditioning is measured by the ratio b/a (close to 1 = well-conditioned; close to 0 = poorly conditioned). For P: b/a = 0.07/0.08 = 0.875 (well-conditioned). For Q: b/a = 0.02/0.20 = 0.10 (poorly conditioned). Point Q's network geometry has a strong directional weakness — likely due to poor azimuthal distribution of observations. Both the shape ratio AND the absolute sizes must be reported for complete accuracy assessment, as required in NAMRIA and DPWH geodetic control specifications.
Wrong Answer
Point P, because its ellipse is nearly circular and therefore shows better survey quality overall.
Correct Answer
Point P has better-conditioned geometry (the ellipse is nearly circular, indicating uniform precision in all directions). Point Q has poorer geometric conditioning (highly elongated ellipse indicating a preferred direction of weakness), even though it is more precise than P in the direction of its minor axis.
Misconception Id
M6
Correct Vs Incorrect
Correct Approach
Survey A: circular, σ = 0.10 m in all directions — geometrically well-conditioned but with moderate absolute uncertainty. Survey B: elongated, σ = 0.01 m perpendicular to the major axis — extremely precise in that direction, but with weak geometry along the major axis direction (σ = 0.15 m). Which is 'better' depends on the purpose of the survey and the direction of interest. For a boundary line running perpendicular to Survey B's major axis, Survey B is far superior.
Incorrect Approach
Survey A has a circular ellipse with semi-axes a = b = 0.10 m. Survey B has an elongated ellipse with a = 0.15 m and b = 0.01 m. Student says Survey A is better because it is 'rounder.' This is wrong — Survey B is far more precise in the direction of its minor axis.
Why Students Believe It
Intuitively, 'round = uniform = good' and 'elongated = stretched = bad' seems like a reasonable visual interpretation. Students conflate the shape of the ellipse (which describes geometric conditioning) with the SIZE of the ellipse (which describes absolute accuracy). The shape-accuracy conflation is intuitive but incorrect.
The matrix formula Σ_y = J Σ_x Jᵀ is just the scalar propagation formula rewritten in matrix form — it gives the same result and is interchangeable.
Tags
- major_error
- matrix_form
- correlation
- covariance_propagation
Topic
Matrix Propagation — Correlated Observations
Severity
major
Exam Impact
Advanced board exam questions involving propagation of uncertainty through a function of adjusted coordinates (which are correlated) test exactly this distinction. A student using only the scalar formula will get the wrong propagated uncertainty when the input covariance matrix has non-zero off-diagonal terms.
The Reality
The scalar formula σ_y² = Σ(∂f/∂xᵢ)²σᵢ² is valid ONLY for independent (uncorrelated) observations. It is a special case of the matrix form where Σ_x is diagonal (all covariances are zero). The matrix form Σ_y = J Σ_x Jᵀ handles the GENERAL case where the input quantities xᵢ may be correlated. In least-squares adjustment, adjusted parameters are almost always correlated (off-diagonal terms in Σ_x̂ are non-zero), so propagating their uncertainty into a derived quantity REQUIRES the full matrix form. Using only the scalar formula when correlations exist will give an incorrect (usually underestimated) propagated uncertainty.
Trap Question
Question
A horizontal distance D is computed from adjusted coordinates as D = √[(ΔE)² + (ΔN)²]. The covariance matrix of (ΔE, ΔN) is Σ = [[0.0009, 0.0006],[0.0006, 0.0016]] m². Is it correct to compute σ_D² = (∂D/∂ΔE)²(0.0009) + (∂D/∂ΔN)²(0.0016) and ignore the off-diagonal element?
Explanation
Since ΔE and ΔN are adjusted coordinates from the same least-squares solution, they are correlated (σ_ΔEΔN = 0.0006 ≠ 0). The scalar formula without the cross-term is only correct for independent quantities. The matrix form Σ_D = JΣJ^T automatically includes the cross-term through the matrix multiplication. Omitting it here will underestimate or overestimate σ_D depending on the sign of σ_ΔEΔN and the partial derivatives.
Wrong Answer
Yes, the formula σ_D² = (∂D/∂ΔE)²σ²_ΔE + (∂D/∂ΔN)²σ²_ΔN is the standard propagation formula and is always correct.
Correct Answer
No. The full propagation must include the covariance term: σ_D² = (∂D/∂ΔE)²σ²_ΔE + 2(∂D/∂ΔE)(∂D/∂ΔN)σ_ΔEΔN + (∂D/∂ΔN)²σ²_ΔN. Ignoring σ_ΔEΔN = 0.0006 m² is only valid if the off-diagonal element is zero.
Misconception Id
M7
Correct Vs Incorrect
Correct Approach
Full propagation: σ_y² = (∂f/∂x)²σ_x² + 2(∂f/∂x)(∂f/∂y)σ_xy + (∂f/∂y)²σ_y². In matrix form: Σ_y = J[[σ_x², σ_xy],[σ_xy, σ_y²]]Jᵀ. The cross-term 2(∂f/∂x)(∂f/∂y)σ_xy captures how correlated errors in x and y jointly affect the derived quantity y.
Incorrect Approach
Given Σ_x̂ with off-diagonal covariance σ_xy ≠ 0, student writes σ_y² = (∂f/∂x)²σ_x² + (∂f/∂y)²σ_y² ignoring the cross-term. This omits the contribution 2(∂f/∂x)(∂f/∂y)σ_xy, which can be positive or negative — a significant systematic omission.
Why Students Believe It
Students learn the scalar formula first and when the matrix version is introduced, they assume it is merely notational shorthand for the same computation. They do not realize the matrix form carries COVARIANCE information between correlated parameters, which the scalar formula (for independent quantities) cannot handle.
A larger weight P assigned to an observation means that observation has a larger standard deviation (is less precise).
Tags
- major_error
- conceptual_gap
- weight_definition
- precision_vs_weight
Topic
Weights in Least-Squares Adjustment — Inverse Variance Relationship
Severity
major
Exam Impact
Questions on setting up weight matrices, interpreting relative precision of observations, or understanding why certain observations dominate an adjustment all hinge on this inverse relationship. A student with this misconception will assign weights backwards, producing a completely wrong adjusted solution.
The Reality
Weight P is INVERSELY proportional to variance: P = σ₀²/σᵢ². A LARGER weight means SMALLER variance means HIGHER precision (smaller σ). Conversely, a small weight means the observation is less reliable (larger σ). The weight matrix P is the key input to the normal equations N = AᵀPA, and the resulting covariance matrix Σ_x̂ = σ̂₀²(AᵀPA)⁻¹ — all correctly reflect this inverse relationship. This is one of the most fundamental definitions in least-squares adjustment theory.
Trap Question
Question
Two distance measurements are made. Measurement A has σ_A = 10 mm and Measurement B has σ_B = 5 mm. If the weight of Measurement A is taken as 1, what is the weight of Measurement B?
Explanation
Weight is inversely proportional to variance. Since σ_B is half of σ_A, the variance of B is one-quarter of A's variance, so B's weight is 4 times A's weight. The more precise measurement (smaller σ) receives a LARGER weight — it contributes more to the adjusted result. This is the fundamental principle of weighted least squares, used throughout traverse adjustments, level network adjustments, and GPS baseline processing in NAMRIA surveys.
Wrong Answer
Weight of B = 0.5, because B has a smaller σ (less uncertainty) so it gets a smaller weight.
Correct Answer
Weight of B = σ_A²/σ_B² = (10)²/(5)² = 100/25 = 4
Misconception Id
M8
Correct Vs Incorrect
Correct Approach
Using relative weights with σ₀ = 1: P₁ = 1/σ₁² = 1/25 = 0.04 and P₂ = 1/σ₂² = 1/4 = 0.25. Or equivalently, normalize: P₁ = 1, P₂ = 25/4 = 6.25 (observation 2, being more precise, gets a weight 6.25 times larger than observation 1). The more precise observation carries more influence in the adjustment.
Incorrect Approach
Observation 1 has σ₁ = 5 mm (less precise). Student assigns P₁ = 5 (larger weight to the less precise observation). Observation 2 has σ₂ = 2 mm. Student assigns P₂ = 2. This is the REVERSE of correct — the less precise observation now dominates the adjustment.
Why Students Believe It
The word 'weight' in everyday language means 'importance' or 'heaviness' — a larger weight seems to mean something bigger or more burdensome, possibly more uncertain. Students who have not internalized the formal definition confuse weight with variance.
The semi-axes of the error ellipse are the eigenvalues of the 2×2 covariance matrix.
Tags
- minor_error
- eigenvalue_confusion
- matrix_computation
- units_error
Topic
Error Ellipse — Eigenvalues and Semi-Axes
Severity
minor
Exam Impact
When a board exam provides the eigenvalues of a covariance matrix and asks for the ellipse dimensions, a student with this misconception will report √λ values as λ values (off by a square root), giving semi-axes that are too large by a factor of ~√λ. This causes wrong numerical answers in ellipse-dimension problems.
The Reality
The semi-axes of the standard error ellipse are the SQUARE ROOTS of the eigenvalues of the 2×2 covariance submatrix. If the eigenvalues of Σ are λ₁ ≥ λ₂, then the semi-major axis a = √λ₁ and semi-minor axis b = √λ₂. This is because eigenvalues are in units of m² (variances), while the semi-axes are in units of m (standard deviations / distances). The eigenvectors give the ORIENTATION of the ellipse axes relative to the coordinate axes.
Trap Question
Question
The 2×2 variance-covariance matrix for an adjusted point yields eigenvalues λ₁ = 0.0064 m² and λ₂ = 0.0016 m². What are the semi-axes of the standard error ellipse?
Explanation
The eigenvalues of the covariance matrix have units of m² (they are variances in the principal directions). The semi-axes of the ellipse are standard deviations — distances in meters — so they are the square roots of the eigenvalues: a = √λ₁ = 0.08 m and b = √λ₂ = 0.04 m. Note that for the 95% confidence ellipse, these semi-axes would be further scaled by √5.991 ≈ 2.45: a₉₅ ≈ 0.196 m, b₉₅ ≈ 0.098 m.
Wrong Answer
a = 0.0064 m and b = 0.0016 m
Correct Answer
a = √0.0064 = 0.08 m and b = √0.0016 = 0.04 m
Misconception Id
M9
Correct Vs Incorrect
Correct Approach
Semi-major axis a = √λ₁ = √0.0025 = 0.05 m. Semi-minor axis b = √λ₂ = √0.0009 = 0.03 m. The corresponding eigenvectors give the orientation (azimuth) of the major axis. The area of the standard error ellipse is π·a·b = π(0.05)(0.03) = 0.00471 m².
Incorrect Approach
Eigenvalues of Σ are λ₁ = 0.0025 m² and λ₂ = 0.0009 m². Student reports semi-major axis a = 0.0025 m and semi-minor axis b = 0.0009 m. These are the eigenvalues (in m²), not the semi-axes (in m) — wrong units and wrong magnitudes.
Why Students Believe It
Students vaguely remember that eigenvalues are involved in computing the error ellipse, and since eigenvalues of the covariance matrix have units of m² (same as variances), they think the semi-axes are directly the eigenvalues.
Error propagation only applies when you make a mathematical computation (like computing area). It does not apply to directly measured quantities.
Tags
- minor_error
- conceptual_gap
- measurement_chain
- general_propagation
Topic
Law of Propagation of Variances — General Applicability
Severity
minor
Exam Impact
Questions that ask students to trace the source of position uncertainty in a traverse, GPS network, or level network require understanding that propagation occurs at every stage — not just in a final derived calculation. Misunderstanding this leads to incomplete error analysis answers in problem-solving questions.
The Reality
Error propagation applies whenever uncertain quantities are combined or transformed, including in everyday measurement scenarios like traversing (angles and distances propagate into coordinate errors), levelling (individual rod readings propagate into height differences), and GPS processing (carrier-phase observations propagate into baseline vectors). Even a 'direct' measurement involves a chain of readings (instrument reading, calibration, environmental correction) each with its own uncertainty. Furthermore, in least-squares adjustment, error propagation through the design matrix A and weight matrix P is the mechanism by which observational errors become coordinate uncertainties via Σ_x̂ = σ̂₀²(AᵀPA)⁻¹.
Trap Question
Question
In a closed traverse adjustment, the angular misclosure is distributed equally among all interior angles. After adjustment, the standard error of an adjusted angle is σ_θ = 5″. A student claims that the standard error of the computed traverse closure was determined by error propagation, but the standard error of each directly-measured interior angle is not a result of propagation. Is the student correct?
Explanation
Even a 'single' angle measurement is the result of combining two direction observations. σ_θ² = σ_FS² + σ_BS². If both direction readings have equal σ_d: σ_θ = σ_d√2. This is error propagation at the most fundamental level. The common exam mistake is to treat directly-read angles as having 'original' errors with no propagation history — but in reality, the entire measurement process from instrument reading to final adjusted coordinate is a continuous chain of error propagation.
Wrong Answer
Yes, because individual angle measurements are directly observed, not computed from other quantities.
Correct Answer
No. Each interior angle is typically the difference of two direction readings (foresight minus backsight), so its uncertainty is itself the result of error propagation: σ_θ = √(σ_FS² + σ_BS²). Propagation occurs at every level of the measurement chain.
Misconception Id
M10
Correct Vs Incorrect
Correct Approach
Error propagation is a chain: instrument reading errors → measured angle/distance errors → via the traverse computation → position errors at each station. Σ_x̂ = σ̂₀²(AᵀPA)⁻¹ captures the complete propagation through the entire network adjustment, not just the final step.
Incorrect Approach
Student says: 'The position error in the traverse endpoint comes only from the angle and distance calculations, not from the individual instrument readings.' This ignores that the 'angle measurement' itself is the result of propagating two pointing errors (forward and back sight), and the 'distance measurement' propagates EDM systematic and random components.
Why Students Believe It
The word 'propagation' implies something is being transferred through a calculation. Students think that if they directly measure a distance, there is no propagation — propagation only comes in when they calculate something derived from the measurement. This misses the broader framework.
The a posteriori variance factor σ̂₀² always equals 1 after a correct adjustment.
Tags
- minor_error
- variance_factor
- residuals
- a_posteriori
Topic
A Posteriori Variance Factor — Covariance Matrix Scaling
Severity
minor
Exam Impact
Problems that ask for the covariance matrix of adjusted parameters using the computed a posteriori σ̂₀² (rather than assuming it equals 1) require the student to correctly compute VᵀPV/(n-u) and use it as a multiplier. Using 1 instead of the actual σ̂₀² will give wrong variances for the adjusted coordinates.
The Reality
The a priori variance factor σ₀ = 1 is an assumption, not a result. The a posteriori variance factor σ̂₀² = VᵀPV/(n-u) is computed from the actual residuals of the adjustment. If the observations and their weights are correctly specified, σ̂₀² should be close to 1 — but it is rarely exactly 1 due to random variation. A value significantly different from 1 is a diagnostic: σ̂₀² >> 1 suggests the weights are too large (observations are worse than assumed) or there are blunders; σ̂₀² << 1 suggests weights are too conservative (observations are better than assumed). The covariance matrix of the adjusted parameters is σ̂₀²(AᵀPA)⁻¹, and using σ̂₀² ≠ 1 changes the magnitude of all reported uncertainties.
Trap Question
Question
A least-squares adjustment with n = 8 observations and u = 3 unknowns yields VᵀPV = 12.5. The normal equation matrix inverse is (AᵀPA)⁻¹ = [[0.01, 0],[0, 0.02]] m². What is the standard error of the second unknown parameter?
Explanation
The a posteriori variance factor σ̂₀² = 2.5 (not 1) must be used to scale the covariance matrix. The covariance matrix of adjusted parameters is Σ_x̂ = σ̂₀²(AᵀPA)⁻¹. Ignoring the computed σ̂₀² and assuming it equals 1 gives σ_x₂ = 0.141 m instead of the correct 0.224 m — a 59% underestimate of the true parameter uncertainty. The value σ̂₀² = 2.5 also signals that the actual observation quality is worse than assumed by the weights.
Wrong Answer
σ_x₂ = √0.02 = 0.141 m (using σ̂₀² = 1 assumed)
Correct Answer
σ̂₀² = 12.5/(8-3) = 12.5/5 = 2.5; σ²_x₂ = σ̂₀² × 0.02 = 2.5 × 0.02 = 0.05 m²; σ_x₂ = √0.05 = 0.224 m
Misconception Id
M11
Correct Vs Incorrect
Correct Approach
σ̂₀² = VᵀPV/(n-u) = 8.4/(10-4) = 8.4/6 = 1.4. Σ_x̂ = 1.4·(AᵀPA)⁻¹. All coordinate standard errors are scaled by √1.4 ≈ 1.183 compared to the unit variance assumption. A chi-squared test can be performed to check if σ̂₀² = 1.4 is statistically consistent with the expected value of 1.
Incorrect Approach
After adjustment with 10 observations and 4 unknowns, VᵀPV = 8.4. Student reports σ̂₀² = 1 (assumed) and uses Σ_x̂ = 1·(AᵀPA)⁻¹. This ignores the actual variance factor.
Why Students Believe It
Students are taught that when you set up the weight matrix using σ₀ = 1 (unit weight standard deviation), and the adjustment is correct, everything 'normalizes' to 1. They confuse the a priori value (σ₀ = 1 assumed) with the a posteriori estimate σ̂₀² = VᵀPV/(n-u) computed from residuals.
The orientation (azimuth) of the error ellipse major axis is always aligned with either the North-South or East-West direction.
Tags
- minor_error
- ellipse_orientation
- eigenvector
- covariance_interpretation
Topic
Error Ellipse — Orientation and Eigenvectors
Severity
minor
Exam Impact
Board exam problems that ask for the orientation angle of the error ellipse, or that present a rotated ellipse and ask about its meaning, test this concept. A student who always assumes N-S or E-W alignment will compute the wrong orientation angle and potentially misidentify the direction of maximum positional uncertainty.
The Reality
The orientation of the error ellipse is determined by the EIGENVECTORS of the 2×2 covariance matrix, not by the coordinate axes. When σ_EN ≠ 0 (off-diagonal covariance is nonzero — which is the general case in a correlated adjustment), the principal axes of the ellipse are rotated relative to the E-N grid. The rotation angle θ of the major axis from the E-axis is given by: tan(2θ) = 2σ_EN / (σ²_E - σ²_N). Only when σ_EN = 0 do the ellipse axes align with the coordinate grid axes.
Trap Question
Question
The 2×2 variance-covariance matrix of a survey point in PPCS Zone IV coordinates is: Σ = [[0.0016, 0.0012],[0.0012, 0.0025]] m². Without computing eigenvalues, can you conclude that the major axis of the error ellipse points in the North direction?
Explanation
The diagonal elements of the covariance matrix give the variance in the E and N directions respectively — they do NOT give the orientation of the ellipse axes unless the matrix is diagonal (σ_EN = 0). When σ_EN ≠ 0, the principal directions of uncertainty are rotated and must be found from the eigenvectors. This is why computing eigenvalues and eigenvectors of the covariance submatrix is the correct procedure for determining error ellipse parameters — a fundamental skill in geodetic network quality assessment under NAMRIA standards.
Wrong Answer
Yes, because σ²_N = 0.0025 m² is larger than σ²_E = 0.0016 m², so the major axis points North.
Correct Answer
No. The off-diagonal covariance σ_EN = 0.0012 m² is nonzero, so the ellipse axes are NOT aligned with the E-N grid. The actual orientation requires computing the eigenvectors: tan(2θ) = 2(0.0012)/(0.0016-0.0025) = 0.0024/(-0.0009) = -2.667, giving 2θ = -69.4°, θ ≈ -34.7° from East (approximately NW-SE orientation).
Misconception Id
M12
Correct Vs Incorrect
Correct Approach
Eigenvalues: λ₁,₂ = [(0.0025+0.0036)/2] ± √{[(0.0025-0.0036)/2]² + (0.0015)²} = 0.00305 ± √(0.00000302 + 0.00000225) = 0.00305 ± 0.00230. So λ₁ = 0.00535 m², λ₂ = 0.00075 m². Semi-axes: a = √0.00535 = 0.0732 m, b = √0.00075 = 0.0274 m. Rotation: tan(2θ) = 2(0.0015)/(0.0025-0.0036) = 0.003/(-0.0011) = -2.727, so 2θ = -69.9°, θ = -34.9° from E-axis. The major axis is tilted 34.9° from East, NOT aligned with any grid axis.
Incorrect Approach
Given Σ = [[0.0025, 0.0015],[0.0015, 0.0036]], student reports major axis is along N-axis (because σ²_N > σ²_E) and reports semi-major axis = √0.0036 = 0.06 m. This ignores the off-diagonal covariance and misidentifies both the orientation and the actual semi-axes.
Why Students Believe It
Since the covariance matrix is expressed in the (Easting, Northing) coordinate system of PPCS/UTM, students assume the eigenvalues correspond to the E and N directions respectively, and therefore the ellipse axes align with the grid axes. This is only true when the covariance matrix is diagonal (zero off-diagonal covariance).
Quick Self Check
For independent measurements, errors add in quadrature: σ_sum = √(σ₁² + σ₂²). Linear addition overestimates the combined uncertainty and is only correct for perfectly correlated (ρ = +1) errors.
Statement
The standard deviation of the sum of two independent measurements is equal to the sum of their individual standard deviations.
The total distance is the SUM of n independent segment measurements. By propagation: σ_sum = σ√n. This is different from the standard error of the MEAN of n readings, which is σ/√n.
Statement
If a distance is taped in n equal segments each with standard error σ, the total distance has a standard error of σ√n.
The 68% rule applies in 1D (±1σ interval). In 2D, the standard error ellipse (1σ ellipse) encloses only approximately 39% of the probability. The 95% confidence ellipse requires scaling the semi-axes by √5.991 ≈ 2.45.
Statement
The standard error ellipse encloses approximately 68% of the 2D probability distribution of the adjusted position.
Diagonal elements Σᵢᵢ = σ²(x̂ᵢ) are variances (units: m²). To obtain standard deviations (units: m), take the square root: σ(x̂ᵢ) = √Σᵢᵢ. Off-diagonal elements are covariances.
Statement
The diagonal elements of the variance-covariance matrix Σ_x̂ represent the variances of the adjusted parameters.
Weight is inversely proportional to variance: P = σ₀²/σ². A larger weight means SMALLER variance, hence HIGHER precision. The more reliable observation gets the larger weight and contributes more to the adjusted result.
Statement
In least-squares adjustment, an observation with a larger weight is less precise (has a larger standard deviation) than one with a smaller weight.
Eigenvalues λ₁ ≥ λ₂ of the covariance matrix have units of m² (variances). The semi-axes a = √λ₁ and b = √λ₂ have units of meters. The eigenvectors give the orientation angles of the ellipse axes.
Statement
The semi-axes of the standard error ellipse are the square roots of the eigenvalues of the 2×2 covariance submatrix.
A circular ellipse indicates uniform geometric conditioning (equal precision in all directions), not necessarily high absolute accuracy. A circular ellipse with large semi-axes has poor accuracy; an elongated ellipse with a very small minor axis has excellent accuracy in one direction. Both shape AND size must be assessed.
Statement
A perfectly circular error ellipse always indicates that the adjusted coordinates have better absolute accuracy than an elongated error ellipse.
When inputs are uncorrelated, Σ_x is diagonal (all off-diagonal elements are zero). The matrix multiplication JΣ_xJᵀ then reduces to the scalar sum Σ(∂f/∂xᵢ)²σᵢ². For correlated inputs (Σ_x has nonzero off-diagonal terms), the full matrix form must be used to include cross-covariance contributions.
Statement
The matrix propagation formula Σ_y = JΣ_xJᵀ reduces to the scalar formula σ_y² = Σ(∂f/∂xᵢ)²σᵢ² only when all input variables are uncorrelated.
Ready to practise for the GELE 2026?
Super Tutor's AI review plan adapts to your weak areas and builds a weekly practice schedule around your target GELE exam date.