Gee-gee definition: 1. Horse races where you try to win money by correctly guessing which horse will win. 177.6k Followers, 1,412 Following, 295 Posts - See Instagram photos and videos from Gee Nelly (@geenelly). Come fly with us to our beloved Sri Lanka on wings of song every day here at MIYURU GEE. Here is a library of your favorite Sri Lankan music, to get lost in, to dream with. We will bring you a new composition every week. So remember to visit us often. On July 22, 2020, Global Eagle Entertainment Inc. (“Global Eagle” or the “Company”) announced that it has agreed upon a definitive asset purchase agreement under which substantially all of the Company’s assets will be sold for total consideration of $675 million to holders of approximately 90% of the Company’s senior secured first-lien term loans.
Suppose we observe repeated measurements (responses and/or covariates) on a group of subjects.We’re interested in modeling the expected response for an individual based on these covariates.Some examples might include
- assigning individuals to one of several controlled diets and measuring their cholesterol over time
- studying the relationship of some variable with earnings over time
- determining the effect of having children on a woman’s probability of participation in the labor force
The benefit of having panel data (repeated measurements) like this is that we can control for time-invariant, unobservable differences between individuals.Having multiple observations per individual allows us to base estimates on the variation within individuals.
The easiest way to do answer these questions would be to fit a linear model to the data, where the covariates have an additive effect on the outcome.If the variables follow something other than a linear relationship (e.g. the response of interest is a probability), a generalized linear model (GLM) would be more appropriate.GLMs have the following form:[Y_i = mu_i + varepsilon_i, qquad g(mu_i) = X_i'beta]
where for individual (i), (Y_i) is the response, (X_i) are covariates, (beta) is a vector of coefficients, (varepsilon_i) is a random error term, and (g) is a link function that maps from the set of possible responses to a linear function of the covariates.
To estimate parameters and do inference with a GLM, we must assume that errors are independent and identically distributed.With panel data, this clearly isn’t the case: observations for each individual are correlated.
As we saw in an earlier presentation, one possible solution is to include subject-specific random effects in the model fitting.This method is called a Generalized Linear Mixed Model (GLMM).GLMMs require some parametric assumptions; if you’re like me (Kellie), assuming that everything is Gaussian probably makes you uncomfortable.
Generalized estimating equations (GEE) are a nonparametric way to handle this.The idea of GEE is to average over all subjects and make a good guess on the within-subject covariance structure.Instead of assuming that data were generated from a certain distribution, uses moment assumptions to iteratively choose the best (beta) to describe the relationship between covariates and response.
Warning: Notice that I did not specify the objective of the analysis.The interpretations of the resultingestimates are different (!) for GLMM and GEE.
GEE estimates population average effects.Consider the following two scenarios:
- Scenario 1: You are a doctor. You want to know how much a statin drug will lower your patient’s odds of getting a heart attack.
- Scenario 2: You are a state health official. You want to know how the number of people who die of heart attacks would change if everyone in the at-risk population took the stain drug.
Source: Allison, P. (2009)
In the first scenario, we want to know the subject-specific odds.In the second, we are interested in the prediction for the entire population.GEE can give us estimates for the second, but not the first.
GEE estimates population-averaged model parameters and their standard errors.The assumptions for GEE are similar to the assumptions for GLMs:
- The responses (Y_1, Y_2, dots, Y_n) are correlated or clustered
- There is a linear relationship between the covariates and a transformation of the response, described by the link function (g).
- Within-subject covariance has some structure (“working covariance”):
- independence (observations over time are independent)
- exchangeable (all observations over time have the same correlation)
- AR(1) (correlation decreases as a power of how many timepoints apart two observations are)
- unstructured (correlation between all timepoints may be different)
We need to pick one of these working covariance structures in order to fit the GEE.As with GLMs, GEE is done using a flavor of iteratively reweighted least squares, plugging in the working covariance matrix as a weight.The weighted least squares problems we fit are the eponymous estimating equations.If you’re familiar with maximum likelihood, you can think of this equation as the score function.This function equals 0 at the optimal choice of (beta).
GEE is a semiparametric method: while we impose some structure on the data generating process (linearity), we do not fully specify its distribution.Estimating (beta) is purely an exercise in optimization.
We have to pick the covariance structure in order to estimate (beta), but what if it’s not right?
Since the estimating equations are really based on the first moment, (beta) will be estimated consistently, even if the working covariance structure is wrong.However, the standard errors computed from this will be wrong.To fix this, use GEE with the Huber-White “sandwich estimator” for robustness.The idea behind the sandwich variance estimator is to use the empirical residuals to approximate the underlying covariance.
Why bother specifying the working covariance to begin with?
- Statistical efficiency
- Sandwich robustness is a large-sample property
Should we use sandwich all the time?
No, it is problematic if
- The number of independent subjects is much smaller than the number of repeated measures
- The design is unbalanced – the number of repeated measures differs across individuals
Question: How does Vitamin E and copper level in the feeds affect the weights of pigs?
Data
- weight of slaughter pigs measured weekly for 12 weeks
- start weight (i.e. the weight at week 1)
- cumulated feed intake
Treatments (3x3 factorial design)
- Vitamin E (dose: 0, 100, 200 mg dl-alpha-tocopheryl acetat/kg feed)
- Copper (dose: 0, 35, 175 mg/kg feed)
Source: Lauridsen, C., Højsgaard, S.,Sørensen, M.T. C. (1999).
- Implementation in R:
geepack
Exchangeable Working Covariance
- Computationally simple relative to MLE counterparts.
- No distributional assumptions.
- Estimates are consistent even if the correlation structure is misspecified (assuming that the model for the mean response is correct)
- Likelihood-based methods are not available for usual statistical inference. GEE is a quasi-likelihood method.
- Unclear on how to perform model selection, as GEE is just an estimating procedure. There is no goodness-of-fit measure readily available.
- No subject-specific estimates; if that is the goal of your study, use a different method.
- GEE2: second-order extension
- The GEE version in this presentation is GEE1.
- Idea: use more complex equations to study the covariance
- Alternating Logistic Regression (ALR) (Carey, Zeger, and Diggle (1993)): model an outcome conditional on another outcome
- Idea: use log odd ratios instead of correlations to model associations
- ONLY the first the mean and the covariance matter (quasi-likelihood approach)
- Use a sandwich estimator to guard against covariance mispecification
- Model population-averaged effects
- Useful when the within-subject dependence is unobserved/unknown
- Still assume subject independence (conditioned on covariates)
GEE
- Liang, K., and S. L. Zeger (1986). Longitudinal data analysis using generalized linear models. Biometrika, 73:13–22.
- Fitzmaurice, G. M., Ware, J.H. and Laird, N. M. (2004). Applied Longitudinal Analysis. Wiley. (Chapter 13)
- Molenberghs, Geert and Verbeke, Geert (2005). Models for Discrete Longitudinal Data. Springer. (Chapter 8)
To GEE or not to GEE:
- Hubbard, A.E., Ahern, J., Fleischer, N.L., Van der Laan, M., Lippman, S.A., Jewell, N., Bruckner, T., Satariano, W.A. (2010). To GEE or not to GEE: comparing population average and mixed models for estimating the associations between neighborhood risk factors and health. Epidemiology 21:467–474.
“…We conclude that the estimation-equation approach of population average models provides a more useful approximation of the truth.”
Subject-specific versus Population-averaged:
Allison, P. D. (2009). Fixed Effects Regression Models (Quantitative Applications in the Social Sciences). SAGE.
Blog post:
Dealing with ugly data: Generalized Estimating Equations (GEE) by BOUSTERHOUT:https://wildlifesnpits.wordpress.com/2014/10/24/dealing-with-ugly-data-generalized-estimating-equations-gee/
Dataset:
Lauridsen, C., Højsgaard, S.,Sørensen, M.T. C. (1999). Influence of Dietary Rapeseed Oli, Vitamin E, and Copper on Performance and Antioxidant and Oxidative Status of Pigs. J. Anim. Sci.77:906-916
Available in the R package geepack
Also found in: Thesaurus, Acronyms, Idioms, Encyclopedia, Wikipedia.
gee 1
(jē)gee 2
(jē)interj.gee 3
also jee(jē)interj.gee 4
(jē)gee 5
(jē)n.gee
(dʒiː) interjgee
(dʒiː) interjGee
(dʒiː) ngee1
(dʒi)interj., v. geed, gee•ing.interj.
gee2
(dʒi)interj.
gee4
(dʒi)n.
gee
Past participle: geed
Gerund: geeing
Imperative |
---|
gee |
gee |
Present |
---|
I gee |
you gee |
he/she/it gees |
we gee |
you gee |
they gee |
Preterite |
---|
I geed |
you geed |
he/she/it geed |
we geed |
you geed |
they geed |
Present Continuous |
---|
I am geeing |
you are geeing |
he/she/it is geeing |
we are geeing |
you are geeing |
they are geeing |
Present Perfect |
---|
I have geed |
you have geed |
he/she/it has geed |
we have geed |
you have geed |
they have geed |
Past Continuous |
---|
I was geeing |
you were geeing |
he/she/it was geeing |
we were geeing |
you were geeing |
they were geeing |
Gee Your Hair Smells Terrific Shampoo
Past Perfect |
---|
I had geed |
you had geed |
he/she/it had geed |
we had geed |
you had geed |
they had geed |
Future |
---|
I will gee |
you will gee |
he/she/it will gee |
we will gee |
you will gee |
they will gee |
Future Perfect |
---|
I will have geed |
you will have geed |
he/she/it will have geed |
we will have geed |
you will have geed |
they will have geed |
Future Continuous |
---|
I will be geeing |
you will be geeing |
he/she/it will be geeing |
we will be geeing |
you will be geeing |
they will be geeing |
Present Perfect Continuous |
---|
I have been geeing |
you have been geeing |
he/she/it has been geeing |
we have been geeing |
you have been geeing |
they have been geeing |
Future Perfect Continuous |
---|
I will have been geeing |
you will have been geeing |
he/she/it will have been geeing |
we will have been geeing |
you will have been geeing |
they will have been geeing |
Past Perfect Continuous |
---|
I had been geeing |
you had been geeing |
he/she/it had been geeing |
we had been geeing |
you had been geeing |
they had been geeing |
Conditional |
---|
I would gee |
you would gee |
he/she/it would gee |
we would gee |
you would gee |
they would gee |
Past Conditional |
---|
I would have geed |
you would have geed |
he/she/it would have geed |
we would have geed |
you would have geed |
they would have geed |
Gee
Noun | 1. | gee - a unit of force equal to the force exerted by gravity; used to indicate the force to which a body is subjected when it is accelerated g-force, g force unit - a unit of measurement of physical force |
Verb | 1. | gee - turn to the right side; 'the horse geed' turn - change orientation or direction, also in the abstract sense; 'Turn towards me'; 'The mugger turned and fled before I could see his face'; 'She turned from herself and learned to listen to others' needs' |
2. | gee - give a command to a horse to turn to the right side cry out, exclaim, call out, outcry, shout, cry - utter aloud; often with surprise, horror, or joy; '`I won!' he exclaimed'; '`Help!' she cried'; '`I'm here,' the mother shouted when she saw her child looking lost' |
gee
1[dʒiː]EXCL (esp US) → ¡caramba!gee whiz! → ¡córcholis!
Geektyper
gee up! → ¡arre!gee
Geeksforgeeks
interjWant to thank TFD for its existence? Tell a friend about us, add a link to this page, or visit the webmaster's page for free fun content.
Link to this page: