The big ideas

The whole course so far on one page: 18 ideas, one sentence each

Start here, or come back here before the exam. Each idea is one sentence in plain language; the link takes you to the section that explains it. If you can say all 18 in your own words, you have the core of Sessions 1–4.

flowchart LR
  A["Spatial data<br/>information + location<br/>(S1)"] --> B["Map rates,<br/>not counts<br/>(S2)"]
  B --> C["Small areas give<br/>unstable rates<br/>(S2)"]
  C --> D["Smooth: borrow<br/>from other areas<br/>EB (S2)"]
  D --> E["Bayesian disease mapping<br/>BYM / BYM2<br/>(S4)"]
  B --> F["Are neighbours alike?<br/>define W, Moran's I, LISA<br/>(S2)"]
  F --> G["Regression with<br/>spatial dependence<br/>SAR / SEM (S3)"]
  F --> E

Lecture map: each arrow means “this leads to the next question”. Not causal. How the four sessions connect.

Session 1 · What spatial epidemiology is

#	Idea	Section
1	Spatial data = information + location, stored as shapes (vector) or grid cells (raster).	1.1
2	Coordinates only make sense with a coordinate reference system; layers must share one before they are combined.	1.3
3	Studies can describe, analyse or intervene; the field mostly does the first two, and a map should start the question, not end it.	1.7
4	MAUP: change the areas (size or borders) and the results can change, even with the same people.	1.10
5	Ecological fallacy: a pattern between areas need not hold for the individuals in them.	1.11
6	Dependence = neighbours are alike; heterogeneity = the relationships themselves differ by place.	1.12

Session 2 · Maps and patterns

#	Idea	Section
7	Counts show where people live, rates show risk, and rates from small populations are unstable.	2.4
8	SMR = observed ÷ expected: above 1 means more cases than the reference risk predicts.	2.5
9	Smoothing pulls noisy small-area rates toward a reference; the less data, the stronger the pull.	2.6
10	Before measuring similarity you must define neighbours (rook, queen, distance, k-nearest) in a weights matrix W.	2.8
11	Moran’s I says whether neighbours are more alike than chance, for the whole map, tested by shuffling the values.	2.11
12	LISA says where: hot spots (HH), cold spots (LL) and outliers (HL, LH).	2.13

Session 3 · Regression with spatial data

#	Idea	Section
13	If residuals cluster on the map, ordinary regression has missed something spatial and its standard errors are too small.	3.1
14	SAR = outcomes spread between neighbours (contagion); SEM = errors share a hidden cause (shared weather). Choose by mechanism.	3.8
15	In SAR a change spreads and feeds back, so report direct, indirect and total impacts, not just β.	3.5

Session 4 · Bayesian disease mapping

#	Idea	Section
16	Pick the likelihood from the outcome and denominator: Binomial for prevalence, Poisson for incidence or observed vs expected.	4.2
17	BYM2 = a spatial part (borrow from neighbours) + a non-spatial part (area’s own), with ρ as the spatial share and τ for the total amount.	4.12
18	Report risk and its uncertainty: a higher estimate is not stronger evidence; look at credible intervals and exceedance probabilities, and remember that a high-risk area is not a cause.	4.17

How to use this page

Cover the “Idea” column and try to say each one from its number and section title. Then open the dashboard to see which sections you have not yet marked as confident.