X1 = B
For every 1 unit increase in X1 the odds of the outcome decrease by 0.3%. You can change the number of units to perhaps make this more meaningful.
What does the distribution of X1 look like? Can you upload a histogram of the variable?
Hi there
i ran logistic regression and got the following model:
var1:
B = -5.670
Sig = 0.01
Exp (B) 0.003
Constant:
B = -0.469
Since the B is negative, is it true to say that for every one unit increase in var1 there is a 99.7 (1-odds ratio) chance of a decrease in the outcome (in this case - no)?
var1 is expressed in proportions (0-1). what is considered one unit here?
thanks very much in advance!
K
X1 = B
For every 1 unit increase in X1 the odds of the outcome decrease by 0.3%. You can change the number of units to perhaps make this more meaningful.
What does the distribution of X1 look like? Can you upload a histogram of the variable?
Stop cowardice, ban guns!
kasperski (08-02-2016)
Thanks so much!
Well, what I was looking for was where the values were landing between 0-1. When using a proportion it is bounded, so say you had most all values at around 90% or higher, interpretations are highly compromised along with confidence intervals. If your histogram is supposed to represent percentages you are probably fine putting var1 in the model, based on my limited knowledge.
Many programs let you set the value of units. If you don't know what the default is in your program, you can sent it to a value and see if the estimates are the same in both models.
Side note, if the program is using 1.00 as the unit and the variable is formatted as percent, then your results seem miniscule regardless of the p-value (e.g., for a 100% increase odds go down 0.3%. Do you get what I am saying?
Stop cowardice, ban guns!
kasperski (08-02-2016)
totally get what you are saying. thanks again for your help!
I hate to bother you again, but i can't figure out what is the default value in SPSS or how it can be set...
thanks again
How is you variable formatted? If in decimals, you could reformat it to integers (multiple by 10) and see if estimates change. I think that would work??
Stop cowardice, ban guns!
kasperski (08-03-2016)
well, when i turn the proportion to percentage (X100) the B changes to -0.057 and Exp(B) changes to 0.945.
that is, the B went from -5.7 to -0.057.
Well I usual just talk out my of rear, but mathematics prevailed. Now comes the question of which one do you use.
Unfortunately I appear to have been a bonehead before. The original interpretation was for a 100% unit increase in X1 the odds of outcome are 99.7% lower. Now we have for a 1% increase in X1 the odds of outcome are 5.5% lower. Do these jive with your hypothesized changes. Another option is to switch the outcome group so you predict the other group then you can say for every 1% increase in X1 the odds of other outcome are 5% (e.g., 1/0.9445)greater.
I guess you get what you pay for
Stop cowardice, ban guns!
kasperski (08-03-2016)
that makes a lot of sense! i can't thank you enough!
No problem. I use logistic reg, kind of regularly, so these questions help me make sure I don't mess up my own analytics - just everybody else's.
Side note, you did not have any proportions in the lower or upper areas of 0-1. So you are limited when making statements to those values that were in your dataset, so say you can't extrapolate to 99%. Which should be intuitive, but we all forget things. Reason being you don't know if those values may have had a different effect in the model (not a linear relationship between 1-99 when using 1 unit increases). Does that make since (perhaps if you fall down 5% of the time it effects your outcome such and such way, but if you fall down 60% of times that 1 unit change isn't the same related to the outcome (you can't just generalize the !% increase because this people have a different odds of outcome). You can create other estimates based on data though, so you can plug values in your model equation and find the odds ratios for certain values say those with X1 = 20 versus 35.
Stop cowardice, ban guns!
well i did have .97, .88 and .85 and 22 more data point under .07, but i get what you are saying.
thanks again!
X1 = B For every one unit increase in X1 odds of a reduction of 0.3% results. You can change the number of units to be able to make this more meaningful. What does the distribution of X1 like? You can upload a graph of the variables?
kasperski (08-08-2016)
thanks for your reply. i have uploaded a histogram earlier in the thread.
Tweet |