Help in explaining my data, lung cancer- air pollution

Hi, i am doing a ecological study of 402 distrcits from accross the UK to se wether Air pollution effects lung cancer mortalilty (measured in SMR)

my variables are
 Low skilled workers – The percentage of houseshold where teh head of the house was employed in in a social class 4 (semi skilled) or social class 5 (unskilled). This is used a indicator of the percentage of population with a low income.
 Ethnicity – Percentage of districts residents from the new comenweatlth and Pakistan. Tkane from the 1991 Census.
 Limiting longterm illness – Percentage of the district who report suffering froma long term illness, data from 1991 census. This is used as ann indicator of general health of teh local population.
 Population density – Persons per km2. Taken from teh Census.
 Particulate Pollution – Average population dose os particulate pollution (PM10) ug m-3 . PM10 particles are very fine particles produced by diesel engines.
 Smoking -percentage of people who smoke.

When induvidually correlated with SMR of lung cancer,
Variable R R2 p(sig)
Air pollution (pm10) 0.413 0.171 <0.01
Social class 0.326 0.106 <0.01
Ethnicity 0.216 0.047 <0.01
Long term ilnes 0.449 0.202 <0.01
Population Density 0.446 0.199 <0.01
Smokers 0.645 0.416 <0.01

As you can see (excluding smoking) Long term ilness and population come up the most sig nif with the highest corelation coeficient.


When peforming a linear regression analysis with all the variables inluded (excluding smoking as i am more interested in the others) Air pollution becomes the most significant (or second when smoking is left in) See table below

Coefficients Coefficients Sig.
B Std. Error
Air pollution (pm10) 3.191 .428 .468 .000
Social class 1.155 .250 .229 .000
Ethnicity -1.126 .357 - .176 .002
Long term ilnes 2.632 .394 .335 .000
Population Density .007 .002 .228 .001

So has anyone got any ideas to why when induvidually correlated, air pollution is not the most sig nif, however when using regression and controlling for others it is.

any ideas/help greatly appreciated.

Many thanks an advance,
Look at the intercorrelations of your predictors. Because your predictors are all likely related to each other, the variable that makes the biggest unique contribution is the one that will stick out.