bear66
Well-known member
No. They're of three types and are single strand but not enveloped as flu/coronaviruses.
Roadmap may be affected - Indian variant
- Thread starter Gazza414
- Start date
SmallTown
Well-known member
Falling for what?
Laughing
Well-known member
To be fair, it's not that obvious sometimes.Chimps and humans share 98.8% of their DNA. Surprising how often it shows.
So pleased your operation was successful. Laughing.
Laughing
Well-known member
Thanks Ziggy.
Colin Warnek
Well-known member
The last year or so in a cartoon.
Boro in Devon
Well-known member
Like it. Unfortunately the models are made by statisticians and are therefore an approximation, so will never give an exact figure.
It is even better when you build a suite of models all using the other models outputs. I used to do this as a job, not in a clinical sense, and you could get the model to say whatever you wanted it to say.
bear66
Well-known member
I am suspicious of the group test meta-analysis stuff. It's very convincing till you find out you can get any answer you want!
Boro in Devon
Well-known member
My world was in banking and finance, I hope clinical data is more heavily peer tested.
You can influence results easily by data cleansing, treatment of outliers, observation and outcome period and target variable definition.
In my opinion it is best to use clean data to build a week model, than dirty data to build a strong model. Dirty data means that you can predict rubbish with a high degree of confidence. Something that was lost on the new graduates that I had to train.
bear66
Well-known member
That's really a more focused multi-variable analysis, but the benefits of multi-variable analysis is the weeding out of variables with weak correlations, so why try to get ahead of the game by making unnecessary assumptions. We'd have stopped all analysis of data to do with covid except that pertaining to old people if you took Bayes to the extreme . . . and learnt nothing.
Laughing
Well-known member
Thats a very good point when looking at the risk facyors. I was more thinking for forecasting infection rates rather than the risk factors which are fairly well understood.
I would have thought a bayes net would work very well with different levels and aspects of lockdown, for example. I only ever hear about it in machine learning and never in statistical analysis.
bear66
Well-known member
I think the difficulty is that the current view is the Indian variant is more transmissible . . but is that due to physics or virology? Bayes could only look at those living in close proximity or those with specific ethnic risk. Or those can be related in which case, don't throw out any data.
In terms of lockdown, ultimately, the quantity and quality of data for risk to health can't be great in making good predictions. I'd like to see the statistical analysis behind the decision to second dose over 50s rather than first dose over 18s!
In terms of predictive models, even 12 months down the line, they are probably good in looking at possible trends but the computing level is more abacus than mainframe computer.
Boro in Devon
Well-known member
When building finance risk models, there were two different schools of thought. The traditional way is to step in the most powerful variables first and then look for uplift for other variables. This would over predict on age and maybe add ethnicity.
The alternative way is to use a stepwise methodology, this would enter weaker variables into the model first and then add more powerful variables afterwards, this usually dealt with correlations and gave a more rounded model better for applying to a population.
Agree, but it’s even more awful that this potential scenario was avoidable. Again. Looks like we are going to have to do it all again for a fourth time.
the_holgate_roof
Well-known member
All because Johnson was after some sort of trade deal....