Trends

[sim khan]

Dear Stata users,

I am trying to fit a trend line to calculate the percentage change for some mortality data (see table below) which I got from Cancer Research UK. I am a little confused on the method. I am assuming that I would use regress with weighted (analytic weights [aweight in stata] ) because mortality increases with age group. I have also heard of another command in stata vwls. Or I could use spline because this dataset curves and you could fit in two lines. I am abit rusty on my regression, so if they are all wrong b=please could you let me know and I wills tick to normal regression (if that is the correct way). I would be greatfull for any advice

Lung Cancer (C33-C34): 2010-2012
Average Number of Deaths per Year and Age-Specific Mortality Rates per 100,000 Population, UK
Age Range	Male Deaths	Female Deaths	Male Rates	Female Rates
0 to 04	0	0	0.0	0.0
05 to 09	0	0	0.0	0.0
10 to 14	0	0	0.0	0.0
15 to 19	0	1	0.0	0.0
20 to 24	2	1	0.1	0.0
25 to 29	3	2	0.1	0.1
30 to 34	9	10	0.5	0.5
35 to 39	34	26	1.7	1.3
40 to 44	103	79	4.5	3.4
45 to 49	246	227	10.7	9.6
50 to 54	557	492	27.3	23.7
55 to 59	1118	945	62.3	51.4
60 to 64	2083	1609	113.7	84.6
65 to 69	2848	2161	188.1	134.5
70 to 74	3404	2468	292.5	189.7
75 to 79	3499	2562	383.9	230.9
80 to 84	2973	2549	479.1	287.5
85+	2560	2572	559.4	271.0
All Ages	19440	15704	62.6	48.8

http://www.cancerresearchuk.org/canc...ity-statistics


Thnak you for your help
Sima

[Nick Cox]

I may be misreading your intent, but I'd suggest that you already have a regression here. A regression is a reduction of the data to a conditional mean function. The data are binned, so unless you have the original data somewhere, you can hardly improve on the regression you have, for several reasons.

Moreover, I would guess that any function defined by an algebraic expression that mimics the data, e.g. being zero for early ages and then taking off *(and then turning down for females), would be too complicated to be useful or informative. Splines are not ruled out, but what would they achieve?

[sim khan]

So would a normal regression analysis suffice? The only problem I have is if I wanted to do any prediction work and I use the %change then wouldn't this be incorrect as we don’t have a straight line (please see attached graph of male mortality rates), but a curve. If this makes sense.

Hence, why I thought perhaps spline (http://www.stata.com/manuals13/rmkspline.pdf) would be a better option. I haven’t used this before but my understanding is that it is used in curve fitting. This method consists of straight lines which are connected by join points (knots). Each knot is the estimated location of change in the slope of the trend line.

I want to understand the method to use and why.

Many thanks,
Sima

[sim khan]

P.S Thank you for your reply.

[Nick Cox]

Sorry, but I can't read .docx extensions from here. Please see FAQ Advice on not using MS Excel or Word attachments.

More crucially, I still can't see why you think plain regression, or even regression using splines, could possibly be interesting or even useful here. Hence I leave the thread open to any other comments.

[sim khan]

Thank you for you reply. I may have missunderstood your last message, but what method would you recommend for calculating the avarage percentage change?