Interpretation of Crime Scene Evidence Exam II

They mention only one proposition (the same source proposition) 

In the paper “Cartridge Case and Bullet Comparison: Examples of Evaluative Reporting”, the Swiss lab uses this language:

(1) Observations must be assessed using at least two propositions in order to be balanced. (2) The observations must be evaluated taking into account relevant case information (for example, the time between the shooting and the seizure of the firearm)

Why are definitive conclusions like Identification not balanced? 

what is a correct description of what likelihood ratios represent?

-The probability of the observations given the first proposition, divided by the probability of the observations given the second proposition

 -the ability of the observations to differentiate between the 2 propositions

-they are a measure of belief change

-image/ridge clarity

-quantity of level 2 features (i.e. minutia)

-specificity/rarity of features

-training about grayscale acuity 

-background distortion 

How does a fingerprint examiner determine sufficiency for purposes of comparison?

grayscale acuity

your ability to separate shades of gray and extract ridge detail from a faint or noisy image

p(observations / proposition)

Which is the correct conditional probability that goes into either the numerator or denominator of a likelihood ratio?

0.00001

Which likelihood ratio provides the least support for the proposition that the defendant touched a surface?

-This profile provides a billion times more support for the proposition that it came from Mr. Smith than if it came from an unrelated male member of the New Zealand population.

-This profile is one billion times more likely if it came from Mr. Smith than if it came from an unrelated male member of the New Zealand population.

Which of the statements are correctly stated when it comes to strength of evidence reporting?

-The probability of observing 12 minutiae in correspondence given a mated pair is very high, while the probability of observing 12 minutiae in correspondence given a non-mated pair is very low.

-The support for the same-source proposition is very high, while the support for the different-sources proposition is very low.

Which is an appropriate way to communicate the results of a forensic comparison if you follow evaluative reporting guidelines?

When you try to copy something or write as someone dictates to you, you may not write in a natural manner.

Why are collected documents preferred over dictated documents?

To determine that writer's natural range of variation.

Why would a FDE want more than one example of a writer's written text?

• Because it provides for more detail and therefore more specificity for the FDE process.

• It is harder to make forgeries of the writer's product.

• It provides more features for comparison.

Why do FDEs prefer writing of high complexity?

• age

• emotional state

• illness

Which below are internal (intrinsic) factors? 

• whether the writer was concentrating

• whether the writer has alternative writing styles

• whether the writer is ambidextrous

• whether the writer was trying to disguise or change their handwriting

• whether the writing changed due to physical or mental conditions

Which below are factors to consider when evaluating dissimilarities?

definitive conclusion

2. Strong probability (highly probable, very probable). The evidence is very persuasive, yet some critical feature or quality is missing so that an identification is not in order; however, the FDE is virtually certain that the questioned and known writings were written by the same individual.

Is this a definitive conclusion or a strength of support statement?

whether an examiner will give the same answer at a later date on the same comparison

repeatability

whether another examiner will give the same answer as the first examiner

reproducibility

• A way to assess how a person's writing might change over time.

• A measure of how similar different writings are from one subject

• The same as within-subject varation

What is intrasubject variation?

1. Laypeople did an initial categorization of the samples into potentially similar writers.

2. They used the Forensic Language Independent Analysis System for Handwriting Identification (FLASH ID, Sciometrics LLC) automated handwriting identification system to find similar handwriting.

3. A team of FDEs reviewed the candidate close-nonmate pairings.

(In the Accruacy and reliability of forensic handwriting comparisons paper) How did they chose non-mates, which determine the difficulty of the study and therefore the overall error rate? 

These are definitive conclusions (was written by; was not written by) that contradict ground truth (mated or non-mated).

(In the Accruacy and reliability of forensic handwriting comparisons paper) How did they define an error?

It conveys different strings of conclusions or different perceived ways of evidence. 

(In the Accruacy and reliability of forensic handwriting comparisons paper) Why are they using a five point scale?

93.1%

(In the Accruacy and reliability of forensic handwriting comparisons paper) A positive predictive value is the probability that are given conclusion is correct. What is the posterior predictive value of written responses?

The two samples were written by twins.

(In the Accruacy and reliability of forensic handwriting comparisons paper) Consider figure 3. These two samples were written by different individuals. However, 23 examiners said they were written by the same person. Why was there such a high error rate for the sample?

Not far off; 85% of the time they were within one category, suggesting that the differences were small.

(In the Accruacy and reliability of forensic handwriting comparisons paper) Examiners only reported the exact same conclusion as other examiners 40.4% of the time. However, when examiners disagreed, how different were they?

A distribution or collection of physical features that serve to differentiate one writer from another.

What is a "writer's profile"?

• Topology is a way to distinguish one writer from another, because if both write the same letter but they have different topologies, then this is useful information to distinguish one writer from another.

• Topology is not changed when the grapheme is stretched, moved, rotated, or otherwise distorted.

• Topology is a way of characterizing a letter that is relatively independent of its shape.

• Topology describes the general connectedness of parts of a letter.

How do the authors use topology?

• Graphemes are collections of the pixels of an image of a handwritten document.

• Graphemes may correspond to character types (letters or numerals).

• Graphemes may consist of parts of characters or parts of groups of characters.

• Graphemes can then be “skeletonized” or “thinned” by one of many available algorithms.

Which below is true of graphemes?

• It allows for angle differences to be calculated between nodes that might provide differentiating information between writers.

• It allows for relative distance measurements between nodes that may systematically differ between writers.

• It gives meaning to geometric measurements

• It allows us to compute relative directions between nodes such as north, south, east or west.

Why is it important to do consistent numbering of nodes?

• The orientation of the major axis

• The length of the major axis

• The ratio of the major to minor axis of the ellipse

The "loops" fit to different features of the handwriting might prove useful for distinguishing between writers. Which aspects of loops might be useful?

A system can select the important features for comparing any two writers.

Once we define a set of features, we need to use these features to distinguish between two writers. How do we use the features to differentiate between writers?

• Handwriting depends on the writing instrument and the writing surface.

• Handwriting can be forged

• Handwriting can change over time.

• Although there may be underlying motoric consistencies, external factors such as rushing or concentrating can affect the handwriting.

• People can disguise their handwriting if they don't want to be associated with a text.

What makes handwriting identification as a discipline hard? 

Laypeople with strong abilities with facial recognition

What are superrecognizers?

They both involve combining several judgments together to improve overall performance.

What do wisdom-of-the-crowds and human-machine fusion have in common?

on a scale from -3 to +3, where -3 is the most support for the different person proposition, and +3 is the most support for the same person proposition.

How did the facial comparison human experts respond during the “Face recognition accuracy of forensic examiners, superrecognizers, and face recognition algorithms” test?

• They were grouped into pairs by the researchers as opposed to being an open search.

• They were chosen to be highly challenging.

• They included mated and nonmated pairs.

How were images chosen for the “Face recognition accuracy of forensic examiners, superrecognizers, and face recognition algorithms” test? 

Examiners

Consider Figure 2. of the “Face recognition accuracy of forensic examiners, superrecognizers, and face recognition algorithms” test. In this graph, the vertical axis is a measure of how good a person is. AUC means area-under-the-curve, and it comes from signal detection theory. Higher values are better. The red dots are the median values for each group.

Which group performed best?

around 3 to 4

Consider the top panel of Figure 3. of the “Face recognition accuracy of forensic examiners, superrecognizers, and face recognition algorithms” test. Fusing is a way of combining across data from multiple subjects. In this case, we take the value from one examiner (on the -3 to 3 scale) and average it with one or more value from other examiners. The resulting average is treated as the response for the group.

The red dot on the far left in Figure 3 is the same dot as the median for Examiners in Figure 2 because we are only "fusing" one person. When the data from multiple examiners is combined, the median goes up. How many examiners need to be combined in order to get essentially perfect performance for the median?

One Examiner and algorithm 2017b

Consider Figure 4. of the “Face recognition accuracy of forensic examiners, superrecognizers, and face recognition algorithms” test. The researchers are fusing people and computer matching algorithms. The red dots are the median performance and are a good summary of the performance of that combination. Which combination proved to give the best performance overall?

• Test video images

• Test using lower-quality images

• Test more recent facial comparison algorithms because they have improved a lot due to AI

• Test different demographic categories

• Test faces from different orientations/viewpoints

What is a limitation of the “Face recognition accuracy of forensic examiners, superrecognizers, and face recognition algorithms” study that should be addressed in future studies?

AFTE Theory of Identification

identifications are made when unique surface contours show "sufficient agreement" exceeding known differences between different tools

• framework requiring evaluation against at least two propositions, incorporating relevant case information

• Reports findings as a numerical Likelihood Ratio

• The forensic observations provide very strong support that bullet B was fired by revolver A. (probabilistic)

ESC, Switzerland in “Cartridge Case and Bullet Comparison: Examples of Evaluative Reporting”

• nine-step logarithmic scale in 2004, ranging from strong support for one proposition to strong support for the alternative

• Avoids the term “Likelihood Ratio” in public reports due to its probabilistic complexity and lack of a good Swedish equivalent; instead uses “Value of Evidence.”

• The evidence provides extremely strong support (Grade +4) that bullet B was fired from firearm A.

NFC, Sweden in “Cartridge Case and Bullet Comparison: Examples of Evaluative Reporting”

• Adopted a Bayesian reporting format with a verbal five-step scale in 2010, later expanded to a seven-step scale in 2015 aligned with ENFSI guidelines

• Uses categorical statements (e.g., “can be excluded”) only when exclusion is definitive, such as mismatched caliber

• The findings are far more probable if bullet B was fired by firearm A than by another similar firearm (evidential strength)

NFI, Netherlands in “Cartridge Case and Bullet Comparison: Examples of Evaluative Reporting”

• balanced

• logical

• robust

• transparent

Four Key Reporting Requirements (ENFSI Guideline)

helps to keep the pellets grouped longer once they leave the barrel

choke

a piece of desensitized photographic paper is treated with a mixture

of sulfanilic acid in distilled water and alpha-naphthol in methanol

Modified Griess Test

• molten metals

• burned gunpowder flakes

• smoke

• microscopic debris

What is in gunshot residue?

• subdural hematoma

• retinal hemorrhages

• encephalopathy

shaken baby syndrome has traditionally been diagnosed based on a “triad” of medical findings

• tested 75 experts by showing them 192 bloodstain patterns and asking for their opinions

• About 11% of their conclusions were wrong when the true cause was known

what was tested in the “Accuracy and reproducibility of conclusions by forensic bloodstain pattern analysis” study?

swinging mechanism

cast-off pattern (BPA- spatter)

blood in airway- breathing, coughing, etc.

expiration pattern (BPA- spatter)

striking an area with liquid blood

impact pattern (BPA- spatter)

breach to circulatory system- hydraulic

projected pattern (BPA- spatter)

gun mechanism

backspatter pattern (BPA)

gravity is the mechanism

drip (BPA)

something wet with blood contacts a non-bloody substrate

transfer stain (BPA)

movement across a surface with wet blood

swipe (BPA- transfer)

pre-existing blood present on a surfface is altered by movement through that blood

wipe (BPA- altered)

an accumulation of blood on a non-porous target surface

pool (BPA- altered)

accumulation of blood on an absorbent material

saturation stain (BPA- altered)

an area where bloodstains would be expected, in an otherwise continuous pattern, and are not present because something has blocked the bloodstains from reaching the target surface

void (BPA- altered)