In this webinar, NewMR Founder Ray Poynter discusses how to look for patterns in research data, as part of his wider 'Finding and Communicating the Story' series.
View the recording via NewMR.org
3. The Framework
1. Define the Problem
2. Assess the Wider Context
3. Find the Big Picture
4. Extract the Key Findings
5. Determine the Message
6. Create the Story
7. Communicate the Story
8. Follow Up
4. 1 Define the Problem
If you don’t define the problem properly, you are unlikely to find the answer
The process includes:
• What is the business question?
• What are the research questions?
• What do we already know?
• What does success look like?
• What does the business plan to do after receiving the answers?
• What are the predictions?
5. Finding the
Patterns in the
Data
1. Using the question as a lens
2. Making the patterns easier to see
3. Look for Connections, Correlations,
Contradictions, Curiosities and
Surprises
4. Assemble the story
6. Making the Patterns Easier to See
• Making numbers more visual
• Sorting
• Comparing
• Benchmarks
• Derived variables
• Analytics
7. Per Capita Govt. Spend on Health - $
183 Countries, 1995 to 2010, source Gapminder
8. 3 Significant Digits – Units of $10
Divide by 10, display no decimals (but don’t delete them)
Three digits good
Two digits medium
One digit poor
9. Sorting – Filters in Excel
Don’t look at data in alphabetical order or in questionnaire order
Pattern 1, Best = USA + Northern Europe
Pattern 2, Worst = Africa overrepresented
Surprise = Niue (Pacific island, pop < 2,000, it’s an outlier)
10. Looking at the Shape of the Data
Steps indicate
possible curiosities,
e.g. OECD, EU etc
11. Index, 1995 = 100
Pattern 1, the most improved tended to
have low incomes per person in 1995
Pattern 2, those who declined tended to
have low incomes per person in 1995
21. Tidy up the Data
% Agreeing Sexy Stylish Traditional Strong Friendly Cheap
Brand 4 97 94 98 97 20 10
Brand 2 87 65 61 96 40 22
Brand 5 93 88 27 30 35 13
Brand 3 72 54 28 28 30 49
Brand 6 31 21 65 75 55 61
Brand 1 42 21 32 42 71 63
Brand 7 11 2 29 60 84 72
Remove the extra columns, multiply by 100 to remove the % signs.
Pattern 1, Brands 4, 2 & 5 are Sexy and Stylish
Pattern 2, Brands 6, 1 & 7 are Friendly and Cheap
22. Do Not (Normally) use the Questionnaire Sequence
0
5000
10000
15000
20000
25000
0
100
200
300
400
500
600
700
Brazil China France Germany India Italy Japan Russia South
Africa
Sweden UK USA World
Cases
Per
Million
Deaths
Per
Million
Deaths and Cases Per Million of Population
(until 27 September 2020)
Deaths / 1Million Case / 1Million
https://www.worldometers.info/coronavirus/ - downloaded 27 September 2020
23. Sort the data by something meaningful
0
5000
10000
15000
20000
25000
0
100
200
300
400
500
600
700
Brazil USA UK Italy Sweden France South
Africa
Russia World Germany India Japan China
Cases
Per
Million
Deaths
Per
Million
Deaths and Cases Per Million of Population
(until 27 September 2020)
Deaths / 1Million Case / 1Million
https://www.worldometers.info/coronavirus/ - downloaded 27 September 2020
World = benchmark
Countries with high death rates
include LatAm, USA, Europe and SA
Lower rates include Europe and Asia
24. Sort the data by something meaningful
0
5000
10000
15000
20000
25000
0
100
200
300
400
500
600
700
Brazil USA UK Italy Sweden France South
Africa
Russia World Germany India Japan China
Cases
Per
Million
Deaths
Per
Million
Deaths and Cases Per Million of Population
(until 27 September 2020)
Deaths / 1Million Case / 1Million
https://www.worldometers.info/coronavirus/ - downloaded 27 September 2020
Some countries break the link
between Cases and Deaths
Idea to test = Were UK, Italy, Sweden
& France testing fewer people?
25. Derived Variables
• Younger people
• People in London
• Men
If a concept is
preferred
(slightly) by
• The differences may be much larger
Create a
variable ‘Young
Men in London’
29. Simplify and Compare
Downloaded from The Guardian 22 Jan 2022
• Pattern = Cases, Hospitalisations, then Deaths
• Alpha had the deaths and hospitalisations
• Delta had the vaccinations and lockdowns
• Omicron had boosters and light restrictions
• The link between cases and deaths changed
37. The Absence of a Strong Pattern Can be Interesting
Source: Statista, 2020
Older people cook
from scratch
more often – but
the differences
between age
groups are not
large.
35-64 almost
identical, 18-24 a
bit lower, 65+ a
bit higher
38. Analytics
We use analytics when the patterns are not visible without
analytics
Key tools include
• Correspondence Analysis – shows relationships on a map
• Factor analysis – great for simplifying the data
• Cluster analysis – shows groupings that exist in the data
• Regression analysis – shows the scale of relationships
• Latent Class – allows techniques to be combined, e.g. regression and
cluster analysis
40. Categorizing the Facts
Fact a Fact d
Fact c
Nice to Know
Fact b
Fact f
Fact g Fact l
Fact e
Fact h
Fact i
Fact j
Fact k
41. Extract the Findings
Fact a Fact d
Fact c
Nice to Know
Fact b
Fact f
Fact g Fact l
Fact e
Fact h
Fact i
Fact j
Fact k
Finding
a
Finding
b
Finding
c
Finding
d
Finding
e
42. Secondary Findings
Fact a Fact d
Fact c
Nice to Know
Fact b
Fact f
Fact g Fact l
Fact e
Fact h
Fact i
Fact j
Fact k
Finding
a
Finding
b
Finding
c
Finding
d
Finding
e
Finding
f
43. Extract the Insight
Fact a Fact d
Fact c
Nice to Know
Fact b
Fact f
Fact g Fact l
Fact e
Fact h
Fact i
Fact j
Fact k
Finding
a
Finding
b
Finding
c
Finding
d
Finding
e
Finding
f
Insight
44. Finding the
Patterns in the
Data
1. Using the question as a lens
2. Making the patterns easier
to see
3. Look for Connections,
Correlations, Contradictions
and Surprises
4. Assemble the story