# Tag Info

11

I'd say logistic regression isn't a test at all; however a logistic regression may then lead to no tests or several tests. You're quite correct that labelling something nonparametric because it's not normal is insufficient. I'd call the exponential family explicitly parametric, so I'd usually regard logistic regression (and Poisson regression and Gamma ...

9

Larry Wasserman defines a parametric model as a set of distributions "that can be parameterized by a finite number of parameters." (p.87) In contrast a nonparametric model is a set of distributions that cannot be paramterised by a finite number of parameters. Thus, by that definition standard logistic regression is a parametric model. The logistic ...

4

There is not enough information here for a definitive answer, but two possibilities are: 1) Adding polynomial terms of X (e.g. $x^2, x^3$) 2) Using some sort of spline curve (e.g. restricted cubic splines) of x. The choice would depend on a few things: 1) How complex is the curve? Very complex relationships are better served by splines. 2) How "explicable" ...

4

A positive interaction effect between A and B means that when A increases the effect (in this case log odds ratio) of B increases and a negative interaction effect means that when A increases the effect of B decreases. When interpreting the results I often find it easiest to work in the odds metric rather than the log(odds) metric. I tend to start with the ...

3

Just use a uniform distribution, or possibly a discrete uniform. You can do this in Excel by using =rand(). (Note that the random number generator that lies behind this function is known to be pretty poor. There are also add-ons that can be purchased that beef up these sorts of capabilities in Excel if you want.) Moreover, you would take the functions ...

2

As regression problems go, it's actually a fairly complicated algorithm. The answer to your question depends a lot on whether you have access to a reliable general-purpose CG optimization algorithm. If you do, the problem becomes somewhat simpler. If you don't, I wouldn't recommend re-implementing logistic regression from scratch (though others have tried, ...

2

To answer you second question: the coefficient is called the log odds ratio, and this is a very apt name as it is literaly the logarithm of the ratio of odds. The odds is the expected number of successes per failure. So if x=1 and and there are $n_1$ persons with x=1 then we expect to find $.05n_1$ successes and $(1-.05)n_1$ failures, so we expect ...

2

There are reasons to include control variables even if they are not significant. E.g. 1) Including them may affect the parameter on the main independent variable (to my mind, this is the true meaning of a "control" variable). 2) Finding a small effect may be important, if others have found a large one. The idea that the null hypothesis always has to be ...

1

As a general principle: 0) you keep the sufficient statistics and the current ML estimates 1) when you get new data, update the sufficient statistics 2) use the previous MLEs as the starting point, use some convenient optimization method to find the new optimum from there. You may need to experiment a bit to find which approaches make the best tradeoffs ...

1

Your first point looks at the "raw data" (confusion matrix: tabulated form) of quality of prediction measurements. There are more options than just misclassification counts: Predictions for different kinds (groups) of cases: training error vs. independent test cases, ... The confusion-matrix based errors (there are also measures here that take into account ...

1

Although a misclassification table sounds intuitive, you need to be very careful in its interpretation as these counts are sensitive to the marginal distribution of the dependent variable. Consider the simplest form a logitstic regression: a logistic regression with just a constant and no explanatory variables. If you are modeling a rare event, say something ...

1

You might use a beta distribution. It has finite support, so that should satisfy the min and max discount constraints that you have. I'm not sure about your $1/N$ requirement, though. You will need to have a discrete distribution if you want to put positive probabilities on the extremes. Maybe you could discretize a continuous distribution, or use an ...

1

I'm not sure if this makes sense but if you were simply solving a gaussian mixture you could use k-means clustering first and estimate an optimal number of components for that. There are many already implemented techniques. Usually, as initial parameters for solving a gaussian mixtures (for example with the EM algorithm), you need an initial guess for the ...

1

Quasi-separation occurs when one cell has a 0 in it. Since the population probability of one cell in the model you show is 0.001, unless the sample is quite large it is likely that this cell will have 0 in the sample. As to your second question - well, that is how coefficients are calculated in logistic regression. See any good book on the subject; e.g. ...

1

First, why is "propensity to dance" binary? That seems like a mistake. I think it would vary along from people with no propensity to dance (e.g. me) to those who will dance at every opportunity, or even make opportunities). But, if it has to be binary, then .... logistic regression is OK here; like other forms of regression, it assumes that there is a ...

1

This rather extensive request for clarification plus the available information I could find will hopefully turn into a more complete answer when we have more information. I see two main possibilities - (A) Plant Growth I found this paper on plant growth: http://www.wseas.us/e-library/transactions/biology/2007/25-645.pdf which suggests that the ...

1

would like to know whether some of the groups of people were more inclined to answer the question with likely than others Well, as you anticipate, you can answer that specific question without going all the way to ordinal logistic regression. If you regard your table "answered 'likely'" vs "answered one of the less-likely categories" (i.e. collapse the ...

1

This can usually be done with the margins command, but gologit2 is an older user-written command that does not take factor variables, so the marginal effects will be off by treating the dummies as if they were continuous. I would estimate your model, use margeff, preserve the data, set the dummies to the base level, set all the continuous variables to zero ...

Only top voted, non community-wiki answers of a minimum length are eligible