11
votes
Accepted
Assumptions of Linear Regression (homoscedasticity and normality of residuals)
The questions themselves are interesting and nontrivial enough that I believe you may have some basic knowledge about assumption testing already, so I'm not telling you what to do in particular (for ...
- 26.9k
6
votes
Is downsampling a valid approach to compare regression results across groups with different sample sizes? If so, how?
It's certainly not a valid way of answering your question.
Your separate analyses show that there is a significant effect in region 1, but no significant effect in region 2.
If your question is ...
- 9,620
5
votes
Is downsampling a valid approach to compare regression results across groups with different sample sizes? If so, how?
I see downsides.
The decrease in data quantity lowers the estimate precision (probably).
There is a further randomness introduced by sub-sampling the sample. At least the data are the data. You can'...
- 65.8k
4
votes
Accepted
What happens to proportional odds ordinal models around 6000 distinct observations?
Yes; nothing special happens at n=6000. You can use rms::orm for as many distinct 𝑌 values as you want. Execution time is roughly linear in the number of distinct ...
- 96.5k
2
votes
Multiple regression correlated predictors
Unfortunately, the short answer is no. The point of multicollinearity is that when two variables are strongly correlated, you have little information with which to differentiate them. As a result, ...
2
votes
Accepted
Is a Direct Relationship Between the Moderator and Dependent Variable Necessary in Moderation Analysis?
I'm not an economist (perhaps terminology is different in that field), but in general Wikipedia says:
the effect of a moderating variable is characterized statistically as an interaction; that is, a ...
- 98k
1
vote
Accepted
If the errors are homogenous but non-normal, can the linear estimator be BLUE?
Let $\mathcal U(\boldsymbol\beta):=\{\mathbf{Ay}+\mathbf a\mid \mathbf{AX}=\mathbf I_p, ~\mathbf a=\mathbf 0\}$ be the class of linear unbiased estimators of $\boldsymbol\beta.$
Under the assumptions ...
- 9,632
1
vote
longitudinal regression but cohort is different in each year?
What you describe just sounds like a cross-sectional design over a time period. A similar approach would be something like an independent samples-same cohort design, or simply a quasi-longitudinal ...
- 16.2k
1
vote
Sample weights in Xgboost regression
This is another question where the easiest solution is the Nike approach: just do it
Here I'm generating some data y that has mean zero and then upweighting the ...
- 43.5k
1
vote
Accepted
Appropriate regression analysis for intervention study using waiting-list control design
While I think this could be handled with a mixed effects model, it might be simpler (from an analysis standpoint) to do the following:
Don't model $T_0$. From the design, we know $E[y \mid t = T_0, ...
- 38.6k
1
vote
Is a Direct Relationship Between the Moderator and Dependent Variable Necessary in Moderation Analysis?
There is no single rule here. Essentially, if M and Y are not correlated, then you are hypothesizing an interaction where the simple slopes intersect somewhere around the mean value of X (perhaps not ...
- 1,929
1
vote
Is there an "observation noise" formulation for logistic regression?
Gaussian random variables have the property that if $X \sim N(0, \sigma^2)$, then $X + \mu \sim N(\mu, \sigma^2$). Hence, we can write $Y = X w^T + \varepsilon$ and this is equivalent to writing $Y \...
- 38.6k
Only top scored, non community-wiki answers of a minimum length are eligible