Presentation is loading. Please wait.

Presentation is loading. Please wait.

Event History Models: Cox & Discrete Time Models

Similar presentations


Presentation on theme: "Event History Models: Cox & Discrete Time Models"— Presentation transcript:

1 Event History Models: Cox & Discrete Time Models
Sociology 229: Advanced Regression Class 6 Copyright © 2010 by Evan Schofer Do not copy or distribute without permission

2 Announcements Assignment 4 Handed out Today’s agenda
More complex EHA assignment Today’s agenda Cox models Parametric Models Reading Discussion

3 Cox Models Where h(t) is the hazard rate
The basic Cox model: Where h(t) is the hazard rate h0(t) is some baseline hazard function (to be inferred from the data) This obviates the need for building a specific functional form into the model Also written as:

4 Cox Model: Example Mostly similar to exponential model…
Cox regression -- Breslow method for ties No. of subjects = Number of obs = No. of failures = Time at risk = Wald chi2(6) = Log pseudolikelihood = Prob > chi2 = (Std. Err. adjusted for 92 clusters in newid3) | Robust _t | Coef. Std. Err z P>|z| [95% Conf. Interval] gdp | degradation | education | democracy | ngo | ingo | Most effects = similar… though education effect loses significance…

5 Cox Model: Baseline Hazard
Cox models involve a “baseline hazard” Note: baseline = when all covariates are zero Question: What does the baseline hazard look like? Or baseline survivor & integrated hazard? Stata can estimate the baseline survivor, hazard, integrated hazard. Two steps: 1. You must ask stata to save the info when you run the Cox model Ex: stcox gdp degradation education democracy ngo ingo, robust nohr basehc(h0) 2. Use “stcurve” command to plot the baseline curves Ex: stcurve, hazard OR stcurve, survival

6 Cox Model: Baseline Hazard
Baseline rate: Adoption of environmental law

7 Cox Model: Baseline Hazard
Note: It may not always make sense to plot the baseline hazard Baseline shows hazard when X variables are zero Sometimes zero values aren’t very useful/interesting Example: Does it make sense to plot hazard of countries adopting laws, if X vars = zero? Hazard rate might be quite low In some cases, you’ll just get a flat zero curve Or extremely high values Solutions: 1. Rescale indep vars before running cox model 2. Use stcurve to choose relevant values of vars.

8 Cox Model: Estimated Hazards
You can also use stcurve to plot estimated hazard rates based on values of indep vars Ex: What is hazard curve if democracy = 1, 5, 10? Strategy: use “at” subcommand: stcurve , hazard at(democ=1) at2(democ=10) NOTE: All other variables are pegged at the mean…

9 Cox: Estimated Hazard Rate
Hazard rate for adoption of environmental law

10 Cox Model Diagnostics Issues that you must deal with:
1. How to estimate results with “ties” in your data Ties = cases that fail at the exact same time 2. How to identify violations of the proportional hazard assumption 3. Dealing with outliers/influential cases 4. Assessing model fit Most of this applies to parametric models Ties are not a concern But, additional issues come up: choosing the right functional form (shape) to model the hazard.

11 Cox Model Issues: Ties How to handle ties in data
It is mathematically complex to estimate models when there are tied failures That is: two cases that have events at the exact same time Several mathematical approaches: Breslow approximation – simplest approach Stata default, but not the best choice! Efron approximation – generally better More computationally intensive, but given the power of modern computers it is not an issue stcox var1 var2 var3, efron

12 Cox Model Issues: Ties Exact marginal – “continuous time approximation” Box-Steffensmeier & Jones: “Averaged Likelihood” Assumes ties didn’t happen EXACTLY at the same time… and considers all possible orderings Exact partial – “discrete” Box-Steffensmeier & Jones: “exact discrete method” Assumes ties happened EXACTLY at the same time Advice: Use Efron at a minimum Exact methods are often more accurate Exact marginal often makes most sense… events rarely occur at the EXACT same time… unless you have discrete data But, exact methods can take a LONG time. For big datasets with many ties, Efron is OK.

13 Proportional Hazard Assumption
Key assumption: Proportional hazards Estimated Hazard ratios are proportional over time i.e., Estimates of a hazard ratio do NOT vary over time Example: Effect of “abstinence” program on sexual behavior Issue: Do abstinence programs lower the rate in a consistent manner across time? Or, perhaps the rate is lower initially… but then the rate jumps back up (maybe even exceeds the control group). Groups are assumed to have “parallel” hazards Rather than rates that diverge, converge (or cross).

14 Proportional Hazard Assumption
Strategies: 1. Visually examine raw hazard plots for sub-groups in your data Watch for non-parallel trends A crude method… not the best approach… but often identifies big violations

15 Proportional Hazard Assumption
Visual examination of raw hazard rate You want them to change proportionally If one doubles, so does the other…

16 Proportional Hazard Assumption
2. Plot –ln(-ln(survival plot)) versus ln(time) across values of X variables What stata calls “stphplot” Parallel lines indicate proportional hazards Again, convergence and divergence (or crossing) indicates violation A less-common approach: compare observed survivor plot to predicted values (for different values of X) What stata calls “stcoxkm” If observed are similar to predicted, assumption is not likely to be violated.

17 Proportional Hazard Assumption
-ln(-ln(survivor)) vs. ln(time) – “stphplot” Parallel=good Convergence suggests violation of proportional hazard assumption (But, I’ve seen worse!)

18 Proportional Hazard Assumption
Cox estimate vs. observed KM – “stcoxkm” Predicted differs from observed for countries in West

19 Proportional Hazard Assumption
3. Piecewise Models Piecewise = break model up into pieces (by time) Ex: Split analysis in to “early” vs “late” time If coefficients vary in different time periods, hazards are not proportional Example: stcox var1 var2 var3 if _t < 10 stcox var1 var2 var3 if _t >= 10 Look for large changes in coefficients!

20 Proportional Hazard Assumption
In a piecewise model, coefficients would differ in non-proportional models Non-Proportional Proportional Early Late Here, the effect is the same in both time periods Here, the effect is negative in the early period and positive in the late period

21 Piecewise Models Look at coefficients at 2 (or more) spans of time
EARLY . stcox gdp degradation education democracy ngo ingo if year < 1985, robust nohr           _t |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]          gdp |           1.05   0.294         degradation |          -1.80   0.072           education |         -0.59   0.552           democracy |           0.87   0.382                  ngo |           2.18   0.029                 ingo |           1.39   0.165        Note: Effect of ngo is larger in early period LATE . stcox gdp degradation education democracy ngo ingo if year >= 1985, robust nohr           _t |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]          gdp |            1.39   0.165          degradation |         -2.38   0.017          education |           0.99   0.323           democracy |           2.59   0.010                  ngo |           1.20   0.229                ingo |         -0.54   0.590       

22 Proportional Hazard Assumption
4. Tests based on re-estimating model Try including time interactions in your model Recall: Interactions – effect of A on C varies with B If effect of variable X on hazard rate (or ratio) varies with time, then hazards aren’t proportional Recall example: Abstinence programs Perhaps abstinence programs have a big effect initially, but the effect diminishes (or reverses) later on

23 Proportional Hazard Assumption
Red = Abstinence group; green = control Positive time interaction No time interaction In non-proportional case, the effect of abstinence programs varies across time

24 Proportional Hazard Assumption
Strategy: Create variables that reflect the interaction of X variables with time Significant effects of time interactions indicate non-proportional hazard Fortunately, inclusion of the interaction term in the model corrects the problem. Issue: X variables can interact with time in multiple ways… Linearly With “log time” or time squared With time dummies You may have to try a range of things…

25 Proportional Hazard Assumption
Red = Abstinence group; green = control Linear time interaction Effect grows consistently over time Try “Abstinence*time” Interaction with time-period… Effect differs early vs. late Try “Abstinence*DLate”

26 Proportional Hazard Assumption
5. Grambsch & Therneau test Ex: Stata “estat phtest” Test for non-zero slope of Schoenfeld residuals vs time Implies log hazard ratio function = proportional Can be applied to general model, or for each variable stcox gdp degradation education democracy ngo ingo, robust nohr scaledsch(sca*) schoenfeld(sch*) . estat phtest Test of proportional hazards assumption Time: Time | chi df Prob>chi2 global test | Significant chi-square indicates violation of proportional hazard assumption

27 Proportional Hazard Assumption
Variable-by-variable test “estat phtest”: Note: Certain variables are especially problematic… . estat phtest, detail Test of proportional hazards assumption Time: Time | rho chi df Prob>chi2 gdp | degradation | education | democracy | ngo | ingo | global test |

28 Proportional Hazard Assumption
Notes on estat phtest : 1. STATA 9/10: Requires that you calculate “schoenfeld residuals” when you run the original cox model And, if you want a test for each variable, you must also request scaled schoenfeld residuals 2. Test is based on identifying non-zero time trend… but how should we characterize time? Options: normal/linear time, log time, time dummies, etc Results may differ depending on your choice Ex: estat phtest, log – specifies “log time” Plot of smoothed Schoenfeld residuals can indicate best way to characterize time Linear trend (not a curve) indicates that time is characterized OK Ex: estat phtest, plot(ngo) OR estat phtest, log plot(ngo)

29 Proportional Hazard Assumption
What if the assumption is violated? 1. Improve model specification Add time interactions to address nonproportionality Ex: If high democracies are not proportional to low democracies, try adding “highdemoc*time” Variables can be interacted with linear time, log time, time dummies, etc., to address the issue 2. Model groups separately Split sample along variables that are non-proportional.

30 Proportional Hazard Assumption
What if the assumption is violated? 3. Use a stratified Cox model Allows a different baseline hazard for each group But, you can’t estimate effect of stratifying variable! Ex: stcox var1 var2 var3, strata(Dhighdemoc) 4. Use a piecewise model Split time into chunks… in which PH assumption is met Requires sufficient sample size in all time periods!

31 Proportional Hazard Assumption
What if the assumption is violated? 5. Live with it (but temper your conclusions) Violation of proportional hazard assumption tends to: Overestimate the effect of variables whose hazard ratios are increasing over time And, underestimate those whose hazard ratios are decreasing However, Allison points out: Cox model is reasonably robust Other issues (e.g., model misspecification) are bigger issues

32 Discrete Time EHA Models
Distinction: Continuous vs. Discrete EHA “Discrete time”: time divided into integer chunks Years, decades, months Spell start & end times are essentially “rounded off” Continuous time: time conceptualized as an unbroken continuum Times need not be rounded off High levels of precision are possible Not just integers, but decimals.

33 Discrete Time EHA Models
Issue: Discrete vs. continuous time gives rise to different EHA models Example: The hazard rate is defined for continuous time: The hazard rate over discrete (identical-sized) chunks of time is (ti):

34 Discrete Time EHA Models
Issue: If the hazard rate in discrete time is a probability, maybe we can model it as such… Standard options for modeling probabilities: Logistic regression (logit) model Probit model Complementary log/log model (cloglog) An asymmetric function Starts slowly from p=0, but accelerates more rapidly toward p=1 at the end Often used when predicted probabilities are very low or high.

35 Discrete Time EHA Models
Example: Discrete time logit model Where p is the probability of an event (Y=1) for a discrete chunk of time Complementary log log model looks like this:

36 Discrete Time EHA Models
Basic logit/probit/cloglog models are like constant-rate/exponential models They assume a constant baseline hazard, represented by constant in the model Discrete EHA models are are proportional hazard models Logit output reports coefficients and odds ratios… But, it is appropriate to refer to them as hazard ratios Coefficient interpretation is the same Raw coeficientss require exponentiation to interpret…

37 Discrete Time EHA: Data
Discrete time models require split-spell data where each spell has constant length Example: every record in your data represents 1 year Number of cases represents total time at risk Ex: If caseid 1 has 10 records, it was at risk for 10 years… This differs from continuous models, where records can represent variable amounts of time E.g., by providing specific start and end times…

38 Discrete Time EHA Data Discrete time data looks like other examples of split spell data But, each record MUST be the same length Example: Country data over time: Logit/probit/cloglog simply models outcome of 1 newname2 newid3 year law eventnum start end ss es pop INDIA INDIA INDIA INDIA INDIA INDIA INDIA INDIA INDIA Event (Y=1)

39 Discrete Time Logit Model
Logit model for discrete time EHA It is a constant rate model In fact, results are almost the same as streg… . logit es gdp degradation education democracy ngo ingo Logistic regression Number of obs = LR chi2(6) = Prob > chi2 = Log likelihood = Pseudo R = es | Coef. Std. Err z P>|z| [95% Conf. Interval] gdp | degradation | education | democracy | ngo | ingo | _cons |

40 Discrete Time and Cox Models
A Cox model can also be estimated in the discrete time context Indeed, the discrete time example helps illustrate what a Cox model really is (even in continuous time) Idea: Use a conditional logit model Conditioned on the cases in the risk set at each point in time … rather than a traditional logit model

41 Discrete Time and Cox Models
A conditional logit model estimates common coefficients across models for many groups Looks at within-group factors, net of overall rate within each group… sorta like a fixed-effects model… Box-Steffensmeier & Jones, p. 80 Thus, effects are modeled net of the “baseline hazard” Interpretation: A Cox model is like pooling a large set of logit results In the continuous time context, the group is the current risk set at the time of any failure

42 Discrete Time and Cox Models
A conditional logit model on discrete time EHA yields identical results to a Cox Model; If you specify the “exact partial” method for handling ties in the continuous time Cox model We’ll cover this later

43 Discrete Time Cox Model
Conditional logit model – a cox model Yields identical results to cox when using discrete data . clogit es gdp degradation education democracy ngo ingo, group(year) Conditional (fixed-effects) logistic regression Number of obs = LR chi2(6) = Prob > chi2 = Log likelihood = Pseudo R = es | Coef. Std. Err z P>|z| [95% Conf. Interval] gdp | degradation | education | democracy | ngo | ingo |

44 Discrete vs. Continuous EHA
In practice, we can often use either discrete or continuous methods Even though time is theoretically continuous, our measures are usually limited to discrete time intervals Ex: year, month, day… For yearly spell data (or any other consistent interval) the data sets are pretty much identical If time resolution is extremely poor, there can be advantages to using discrete time models Otherwise, continuous time models provide greater flexibility And more modeling options.

45 EHA Example In-class group activity: Let’s design a study
Outcome of interest: Students dropping a course What is the risk set? How would you set up the data? What are key independent variables? What kind of model would you use? Work in groups of 2-4, and be prepared to discuss your thoughts…

46 Reading Discussion Empirical Example:  Soule, Sarah A and Susan Olzak.  2004.  “When Do Movements Matter? The Politics of Contingency and the Equal Rights Amendment.”  American Sociological Review, Vol. 69, No. 4. (Aug., 2004), pp Long, J. Scott, Paul D. Allison, and Robert McGinnis.  1993.  “Rank Advancement in Academic Careers:  Sex Differences and the Effects of Productivity.”  American Sociological Review, 58, 5:

47


Download ppt "Event History Models: Cox & Discrete Time Models"

Similar presentations


Ads by Google