Upload
taymour-matin
View
54
Download
1
Tags:
Embed Size (px)
Citation preview
When the fog clears…
a simple truth about data emerges …
There are two types of data…
&
inferred data observed data
One type of inferred data is panel data
Panel data use has proliferated in recent years
0
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
1980 1985 1990 1995 2000 2005 2010 2015 2020
Panel D
ata Stud
ies*
*Social Sciences Citation Index 1986, 2004, and 2005 – keywords “panel data” or “longitudinal data”
Nielsen Homescan Panel
Panel data may be the best option
Panel data is suited to large longitudinal studies
United States • National Longitudinal Surveys of Labor Market
Experience (NLS) • University of Michigan’s Panel Study (MPS)
Europe • Netherlands Socio-‐‑Economic Panel (SEP) • German Social Economics Panel (GSOEP) • Luxembourg Social Panel (PSELL) • British Household Panel Survey (BHS)
Warning: Panel data have limitations
• Panel membership may not be representative • Breaks down at lower levels of aggregation • Low sample size issues are common • Modeling accentuates bias • Panel members may report wrong data • Not timely • Costly
As noted by independent reviews
Nielsen Homescan Data “The different data give different results. Out of the 20 slope parameters, 5 have different signs, 9 do not agree on their statistical significance, and 13 are statistically different.” USDA On the Accuracy of Nielsen Homescan Data
CASE STUDY Measuring Retail Performance
What is the best source for measuring a retailer’s performance?
No surprise, it’s sales!
But, retailers are reluctant to give up sales data
Don’t despair, there’s another way to measure performance
Performance measurement alternatives are within reach
Foot traffic and distance travelled are observable events that can be measured and analyzed
• Deliver timely and relevant marketing offers by offering location-‐‑based ad triggers
• Improve decisions by an improved understanding of customers using time and location analytics
• Rethink competitive strategy based on competitor sets derived from actual visitations
• Enhance financial reporting by benchmarking competitor performance
Benefits of visitation data
Performance management redefined
• Place Visit Rates – PVR • Total site visits • Visits by distance travelled • Visits by time of day • Days since last visit • Visits to top competitors • Profile of best customers • Market share by territory • Competitive benchmarking • Rank of top pre-‐‑visit locations
Performance Management 2.0
Trip Driver Customer Deciles
Trip Driver Model
• The trip driver model predicts future visitations based on historical visitations, average distance traveled, demographic acributes, and other explanatory variables
• Methodologies considered include: generalized linear models (GLM), random-‐‑forest, and other classification models
• The model is validated on a hold-‐‑out sample as well as a test-‐‑sample using rigorous standards to ensure model robustness and conformity to modeling assumptions
The model’s score is grouped into deciles to highlight differences in customer behavior
Trip Driver Model A.ributes 1 2 3 4 5 6 7 8 9 10
Visits per Month 60% of Visits 40% of Visits
Distance Traveled (Miles) < 3 3 -‐‑ 5 > 15
Age 25-‐‑35 35-‐‑45 45-‐‑55
Household Income $85K to $120K $65K to $120K
Grocery Visits per Month < 1.8 > 1.8
Visits to Top 3 Competitors per Month < 2x 2 – 4x > 4x
60% of visits come from 30% of customers
Distance macers
TRIP DRIVER CUSTOMER DECILES
Best customers
skew younger Best
customers skew affluent
Best customers are not active
grocery shoppers Best
customers are loyal
Media Effectiveness
Customer Deciles is overlaid onto PVR lift to identify marketing opportunities
1 2 3 4 5 6 7 8 9 10
Exposed to Ad in Period 1 & Converted in Period 2
Exposed to Ad in Period 1 & Did Not Convert in Period 2
Not Exposed to Ad in Period 1 & Converted in Period 2
Not Exposed to Ad in Period 1 & Did Not Convert in Period 2
PVR Lift
Exposed PVR
Control PVR
TRIP DRIVER CUSTOMER DECILES
For example
CUSTOMER)DECILES)))-
1" 2" 3" 4" 5" 6" 7" 8" 9" 10"
Exposed(to(Ad(in(Period(1(&(Converted(in(Period(24
Exposed(to(Ad(in(Period(1(&(Did(Not(Convert(in(Period(24
Not(Exposed(to(Ad(in(Period(1(&(Converted(in(Period(24
Not(Exposed(to(Ad(in(Period(1(&(Did(Not(Convert(in(Period(24
PVR(Lift4
Exposed(PVR4
Control(PVR4