DIO Oversight: Forensics
During the 20 years that OSI/ORI have existed,
investigators have developed a number of
computer-assisted tools and approaches to help
strengthen institutional findings.
The following slides will provide a few examples
of this.
Detection of Fabricated
Numbers
If sets of transcribed numbers are provided as
“raw data” rather than instrument printouts,
consider whether the numbers might have been
fabricated.
Research by ORI and others shows that
insignificant (right-most) digits in numbers, if real,
e.g. from instruments, are uniformly distributed
while numbers made up by people often are nonuniform.
Here is a DIO scan of a
spreadsheet submitted
by a respondent that
was unaccompanied by
an instrument printout.
Under pressure, he
subsequently provided
similar data that was
accompanied by
printouts from a
scintillation counter.
The tool used by DIO
The DigiProbe Screen
Start with the numbers copied into a text file
Bar graphs effectively illustrate the distribution of digits for
the two right most positions for the data sets without
counter tapes (left) and those with counter tapes
(right).
Data Sets 1-3 (no counter tapes)
Data Sets 4-11
Right-most digit; p = < 0.00001
Second right-most (10s), p = <0.00001
70
60
60
50
50
Frequency
Frequency
40
40
30
30
20
20
Right-most digit; p = 0.778
Second right-most (10s), p = 0.787
10
10
0
0
0
1
2
3
4
5
Digit
6
7
8
9
ORI 4/05
0
1
2
3
4
5
Digit
6
7
8
6
9
DIO analysis of the distribution of the two right most digits in data sets 1-11
(group V)
Probabilities
obtained with
DigiProbe for
the 11 assays
and for the
assays with
and without
counter tapes
grouped
together.
Data set
Total Digits
p value
1
100, 100
<0.00001, <0.00001
2
156, 156
<.00001, 0.00002
3
156, 156
0.02519, 0.00451
4
56, 56
0.57784, 0.47087
5
42, 42
0.22426, 0.22426
6
78, 78
0.46860, 0.57069
7
93, 93
0.27719, 0.97832
8
74, 74
0.07906, 0.51789
9
77, 77
0.33904,0.40441
10
42, 39
0.19768, 0.24219
11
50, 50
0.81654, 0.77919
1-3
412, 412
<0.00001, <0.00001
4-11
512, 509
0.77754, 0.78696
Observations
The most compelling evidence for misconduct in this
case was obtained from the fabricated data sets not
examined by the university.
In no instance was the distribution of non-leading
digits non-uniform when tapes were available, and in
each of the three assays without tapes, digits were
non-uniform at both positions.
This conclusion is made significantly stronger by the
presence of “control data” from companion assays
for which counter tapes were available.
ORI Cases With Questioned Images
80.0
70.0
60.0
50.0
40.0
30.0
20.0
10.0
0.0
1 2 3 4 5 6 7 8 9 10
2 Year Reporting Period ('89-90 to '07-08)
Examples of analyzing images
Several examples follow which illustrate how ORI
can examine images provided by institutions
during their investigation.
Many of ORI’s cases involve images that are
duplicated from paper to paper or paper to grant
application. This may be duplicate publication,
but when such images are said to be the result
of different experiments, one of the images, at
minimum, has been potentially falsified.
The first example, however, is a little different.
ORI 4/05
11
Screen shot from
Photoshop
showing analysis
under way – the
small circle in the
Color Picker is the
brush size moved
to a color
approximately
matching the
image’s
background.
ORI 4/05
12
The result of removing most of the “scribbling.”
ORI 4/05
13
This is why the RIO sent the previous
sample to ORI for Review
In this case, 1 film was used to represent 2 experiments
Corner of Film
the same film used for mouse a and mouse Myo D and Myo G
Scanned film separated by hue
Result: writing in red erased from film then re-labeled
© Copyright 2026 Paperzz