Exploration of SIC Dataset
Histogram of injury rates
A histogram of injury rates was constructed to visualize the distribution of the rates. The distribution is heavily concentrated below a rate of 5 per 100,000 FTE, and quickly thins out as the rate increases. A similar story can be said of the fatality rates, although rates are extremely/heavily concentrated about 0, thinning out immediately after that.
Histogram of fatality rates
These distributions follow a shape similar to that of a poisson distribution; thus, there is justification for constructing poisson models for regression. However, poisson regression requires count data as the dependent variable (but we are modelling rates). As a result, we will model the counts of the injuries and fatalities, while using an offset of the \(\frac{Annual Hours}{200,000}\) (equivalant of 100 FTE) for injuries and \(\frac{Annual Hours}{20,000,000}\) (equivalent to 10,000 FTE). Note: I am once more confused with the FTE and it’s equivalent into worker hours. I would like to figure out if these rates are what they should be…
Let’s visualize what the average noise measurements, average injury rates, and average fatal rates look like over the years. We see that noise measurements in the mining industry have a lot of variability from year to year, although there is a slight decline over time. Injury rates have a real steep and obvious decline, while fatality rates slightly decline.
Below are the injury rates plotted by year and separated by canvass2. Once more, the declines in rates ove time can be seen within each canvass2. There are values higher than 30 per 100 FTE, which are highlighted in red.
Among the fatality rates, less of a trend can be seen due to the overinflation of zeros contained within the dataset, although it still is present by the decline of higher values over time. Values higher than 50 per 10,000 FTE are highlighted in red.
Is there anything to be said about the role of noise exposure to injury and fatality rates? It is a bit unclear by the plots of
What can we say about the impacts of the Hearing Conservation Program on injury rates and noise? Let’s view if there exists a significant difference between injury and fatality rates before 2000 and after 2000.
Overall, we see that the across all canvasses, the injury and fatality rates signficantly dropped. Is the same true for noise exposure?
While there are slight decreases in the average noise exposures before and after the HCP, this difference is not as quite visually obvious. However, a Welch’s t-test of the average noise exposures before and after the HCP indicates a more concrete story.
Percentage of noise measurements larger than 85 dBA before and after the HCP
##
## Welch Two Sample t-test
##
## data: final_twa by HCP
## t = 3.4445, df = 1856.7, p-value = 0.0005849
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## 0.2991018 1.0900954
## sample estimates:
## mean in group <2000 mean in group >=2000
## 83.75211 83.05751
The results of Welch’s t-test among the noise average means indicate that there is a statisticaly signficant lower difference after the HCP was implemented (83.06 dBA after HCP compared to 83.75 prior). Similar tests can be performed on the injury rates and the fatality rates.
##
## Welch Two Sample t-test
##
## data: rate_injuries by HCP
## t = 13.757, df = 1934.1, p-value < 2.2e-16
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## 2.310144 3.078310
## sample estimates:
## mean in group <2000 mean in group >=2000
## 6.926904 4.232678
##
## Welch Two Sample t-test
##
## data: rate_fatal by HCP
## t = 2.8461, df = 1933.6, p-value = 0.004472
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## 0.3009925 1.6351096
## sample estimates:
## mean in group <2000 mean in group >=2000
## 2.123769 1.155718
Yay! For both the average injury rates and the average fatality rates, there was a statistically signficant decrease before and after the HCP was implemented, similarly to the average noise exposures. Could this indicate some level of association between injury/fatality rates and the combination of noise exposures and the impacts of the hearing conservation program? Let’s perform some poisson regression to uncover exactly what effect noise exposure and the HCP had on these incident rates and get some numbers behind these ideas.
Poisson Regression
Due to the collinearity between industries within a canvass, we will be using the industries as a random effect within each of the models. Let us begin with the simplest model: noise exposure on injury rates.
Injury rates
And the fitted v. resid plot is randomly distributed about 0
However, we see that the residuals are not randomly distributed about 0 for both levels of HCP
## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: poisson ( log )
## Formula: n_injuries ~ final_twa + (1 | SIC.AI) + offset(log(offset_injuries))
## Data: .
##
## AIC BIC logLik deviance df.resid
## 59191.4 59208.1 -29592.7 59185.4 1934
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -27.878 -1.896 -0.383 1.762 34.449
##
## Random effects:
## Groups Name Variance Std.Dev.
## SIC.AI (Intercept) 0.1353 0.3679
## Number of obs: 1937, groups: SIC.AI, 80
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -5.6135408 0.0671350 -83.62 <2e-16 ***
## final_twa 0.0868920 0.0006038 143.91 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## final_twa -0.763
## optimizer (Nelder_Mead) convergence code: 0 (OK)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
WOW! The coefficient of noise exposure is statistically signficantally not equal to zero, AND the coefficient is positive? What could be better? The results indicate that a one decibel increase is equal to a 1.02 increase in the incident rate ratio (IRR) of injuries. In other words, a 5 decibel increase in TWA is associated with a 1.09 increase in the IRR of injuries. What if we add HCP as a variable?
And the fitted v. resid plot is randomly distributed about 0
The residuals by HCP are now both randomly distributed about 0 with HCP in the model
## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: poisson ( log )
## Formula:
## n_injuries ~ final_twa + HCP + (1 | SIC.AI) + offset(log(offset_injuries))
## Data: .
##
## AIC BIC logLik deviance df.resid
## 34736.4 34758.7 -17364.2 34728.4 1933
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -23.8424 -1.4605 -0.2966 1.3470 23.0351
##
## Random effects:
## Groups Name Variance Std.Dev.
## SIC.AI (Intercept) 0.1226 0.3501
## Number of obs: 1937, groups: SIC.AI, 80
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.1345997 0.0723334 -15.69 <2e-16 ***
## final_twa 0.0360550 0.0006994 51.55 <2e-16 ***
## HCP>=2000 -0.5463991 0.0035751 -152.83 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) fnl_tw
## final_twa -0.820
## HCP>=2000 -0.390 0.462
## optimizer (Nelder_Mead) convergence code: 0 (OK)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
Once more, we see some great results coming out of the inclusion of HCP in the model. For a one decibel increase in noise measurements, the IRR of injuries increases by 1.01. In other words, a 5 decibel increase in TWA is associated with a 1.07 increase in the IRR of injuries. Futhermore, the IRR descreased by 0.59 for years after the HCP was implemented, as compared to the years preceeding the implementation of the HCP.
Lastly, let us fit canvass to the model to see if there is some sort of interaction with this predictor.
The residuals are still randomly distributed about 0
## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: poisson ( log )
## Formula:
## n_injuries ~ final_twa + HCP + canvass2 + (1 | SIC.AI) + offset(log(offset_injuries))
## Data: .
##
## AIC BIC logLik deviance df.resid
## 34732.6 34766.1 -17360.3 34720.6 1931
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -23.8458 -1.4603 -0.2922 1.3571 23.0306
##
## Random effects:
## Groups Name Variance Std.Dev.
## SIC.AI (Intercept) 0.1092 0.3305
## Number of obs: 1937, groups: SIC.AI, 80
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.6619650 0.2408512 -2.748 0.00599 **
## final_twa 0.0360810 0.0006995 51.583 < 2e-16 ***
## HCP>=2000 -0.5463456 0.0035751 -152.819 < 2e-16 ***
## canvass2Metal -0.6201212 0.2460158 -2.521 0.01171 *
## canvass2Nonmetal -0.4401808 0.2382703 -1.847 0.06469 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) fnl_tw HCP>=2 cnvs2M
## final_twa -0.242
## HCP>=2000 -0.115 0.462
## canvass2Mtl -0.920 -0.008 -0.003
## cnvss2Nnmtl -0.951 -0.003 -0.002 0.932
## optimizer (Nelder_Mead) convergence code: 0 (OK)
## Model failed to converge with max|grad| = 0.0041079 (tol = 0.002, component 1)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
## Model is nearly unidentifiable: large eigenvalue ratio
## - Rescale variables?
Fatality Rates
And the fitted v. resid plot is more or less randomly distributed about 0
## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: poisson ( log )
## Formula: n_fatal ~ final_twa + (1 | SIC.AI) + offset(log(offset_fatal))
## Data: .
##
## AIC BIC logLik deviance df.resid
## 2394.5 2411.3 -1194.3 2388.5 1934
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.8575 -0.4078 -0.1762 -0.0761 12.3827
##
## Random effects:
## Groups Name Variance Std.Dev.
## SIC.AI (Intercept) 0.3899 0.6244
## Number of obs: 1937, groups: SIC.AI, 80
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -5.548906 0.773520 -7.174 7.31e-13 ***
## final_twa 0.071403 0.009054 7.887 3.10e-15 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## final_twa -0.989
## optimizer (Nelder_Mead) convergence code: 0 (OK)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
Even in the case of fatality rates, the coefficient of noise exposure is statistically signficantally not equal to zero. The results indicate that a one decibel increase is equal to a 1.07 increase in the incident rate ratio (IRR) of fatalities. In other words, a 5 decibel increase in TWA is associated with a 1.43 increase in the IRR of fatalities. What if we add HCP as a variable?
And the fitted v. resid plot is randomly distributed about 0
## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: poisson ( log )
## Formula: n_fatal ~ final_twa + HCP + (1 | SIC.AI) + offset(log(offset_fatal))
## Data: .
##
## AIC BIC logLik deviance df.resid
## 2348.8 2371.1 -1170.4 2340.8 1933
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.4889 -0.3886 -0.1769 -0.0765 10.8732
##
## Random effects:
## Groups Name Variance Std.Dev.
## SIC.AI (Intercept) 0.4091 0.6396
## Number of obs: 1937, groups: SIC.AI, 80
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -2.17984 0.92372 -2.360 0.01828 *
## final_twa 0.03316 0.01080 3.071 0.00213 **
## HCP>=2000 -0.38074 0.05564 -6.843 7.78e-12 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) fnl_tw
## final_twa -0.992
## HCP>=2000 -0.523 0.512
## optimizer (Nelder_Mead) convergence code: 0 (OK)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
The model with HCP for fatality rates looks similar to that from injury rates For a one decibel increase in noise measurements, the IRR of fatalities increases by 1.03. In other words, a 5 decibel increase in TWA is associated with a 1.18 increase in the IRR of fatalities. Futhermore, the IRR descreased by 0.68 for years after the HCP was implemented, as compared to the years preceeding the implementation of the HCP.
Lastly, let us fit canvass to the model to see if there is some sort of interaction with this predictor.
The residuals look the same as the other plots
## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: poisson ( log )
## Formula:
## n_fatal ~ final_twa + HCP + canvass2 + (1 | SIC.AI) + offset(log(offset_fatal))
## Data: .
##
## AIC BIC logLik deviance df.resid
## 2347.2 2380.6 -1167.6 2335.2 1931
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.4852 -0.3866 -0.1750 -0.0761 10.7825
##
## Random effects:
## Groups Name Variance Std.Dev.
## SIC.AI (Intercept) 0.3425 0.5852
## Number of obs: 1937, groups: SIC.AI, 80
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.32626 1.00429 -1.321 0.18664
## final_twa 0.03526 0.01090 3.236 0.00121 **
## HCP>=2000 -0.37472 0.05580 -6.716 1.87e-11 ***
## canvass2Metal -1.13874 0.48094 -2.368 0.01790 *
## canvass2Nonmetal -1.05227 0.44539 -2.363 0.01815 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) fnl_tw HCP>=2 cnvs2M
## final_twa -0.905
## HCP>=2000 -0.479 0.516
## canvass2Mtl -0.320 -0.065 -0.034
## cnvss2Nnmtl -0.383 -0.028 -0.019 0.862
## optimizer (Nelder_Mead) convergence code: 0 (OK)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
An interesting result
What if we interact the HCP variable with noise exposure? What can we learn?
## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: poisson ( log )
## Formula:
## n_injuries ~ final_twa * HCP + canvass2 + (1 | SIC.AI) + offset(log(offset_injuries))
## Data: .
##
## AIC BIC logLik deviance df.resid
## 34640.9 34679.9 -17313.5 34626.9 1930
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -24.3229 -1.4758 -0.3026 1.3284 22.7153
##
## Random effects:
## Groups Name Variance Std.Dev.
## SIC.AI (Intercept) 0.1084 0.3292
## Number of obs: 1937, groups: SIC.AI, 80
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.1590917 0.2455638 -0.648 0.5171
## final_twa 0.0301218 0.0009295 32.406 <2e-16 ***
## HCP>=2000 -1.5362498 0.1020103 -15.060 <2e-16 ***
## canvass2Metal -0.6222150 0.2451688 -2.538 0.0112 *
## canvass2Nonmetal -0.4465347 0.2374467 -1.881 0.0600 .
## final_twa:HCP>=2000 0.0119773 0.0012334 9.711 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) fnl_tw HCP>=2 cnvs2M cnvs2N
## final_twa -0.317
## HCP>=2000 -0.215 0.674
## canvass2Mtl -0.899 -0.005 0.001
## cnvss2Nnmtl -0.930 0.000 0.003 0.932
## f_:HCP>=200 0.212 -0.663 -0.999 -0.001 -0.003
## optimizer (Nelder_Mead) convergence code: 0 (OK)
## Model failed to converge with max|grad| = 0.0657948 (tol = 0.002, component 1)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
## Model is nearly unidentifiable: large eigenvalue ratio
## - Rescale variables?
## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: poisson ( log )
## Formula:
## n_fatal ~ final_twa * HCP + canvass2 + (1 | SIC.AI) + offset(log(offset_fatal))
## Data: .
##
## AIC BIC logLik deviance df.resid
## 2345.8 2384.7 -1165.9 2331.8 1930
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.5903 -0.3807 -0.1760 -0.0761 10.8261
##
## Random effects:
## Groups Name Variance Std.Dev.
## SIC.AI (Intercept) 0.3346 0.5784
## Number of obs: 1937, groups: SIC.AI, 80
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -2.87578 1.30507 -2.204 0.027556 *
## final_twa 0.05363 0.01473 3.642 0.000271 ***
## HCP>=2000 2.49610 1.54589 1.615 0.106382
## canvass2Metal -1.11971 0.47605 -2.352 0.018669 *
## canvass2Nonmetal -1.03205 0.44079 -2.341 0.019212 *
## final_twa:HCP>=2000 -0.03476 0.01871 -1.858 0.063217 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) fnl_tw HCP>=2 cnvs2M cnvs2N
## final_twa -0.946
## HCP>=2000 -0.642 0.674
## canvass2Mtl -0.251 -0.040 0.016
## cnvss2Nnmtl -0.304 -0.007 0.023 0.861
## f_:HCP>=200 0.628 -0.660 -0.999 -0.017 -0.023
## optimizer (Nelder_Mead) convergence code: 0 (OK)
## Model failed to converge with max|grad| = 0.0063242 (tol = 0.002, component 1)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
## Model is nearly unidentifiable: large eigenvalue ratio
## - Rescale variables?
Year as a predictor
## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: poisson ( log )
## Formula:
## n_injuries ~ final_twa + Year + (1 | SIC.AI) + offset(log(offset_injuries))
## Data: .
##
## AIC BIC logLik deviance df.resid
## 33278.8 33301.0 -16635.4 33270.8 1933
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -40.393 -1.292 -0.159 1.368 18.760
##
## Random effects:
## Groups Name Variance Std.Dev.
## SIC.AI (Intercept) 0.131 0.3619
## Number of obs: 1937, groups: SIC.AI, 80
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 66.1930991 0.4586966 144.3 <2e-16 ***
## final_twa 0.0195568 0.0007608 25.7 <2e-16 ***
## Year -0.0331355 0.0002093 -158.4 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) fnl_tw
## final_twa -0.656
## Year -0.989 0.568
## optimizer (Nelder_Mead) convergence code: 0 (OK)
## Model failed to converge with max|grad| = 0.0623875 (tol = 0.002, component 1)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
## Model is nearly unidentifiable: large eigenvalue ratio
## - Rescale variables?
## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: poisson ( log )
## Formula:
## n_injuries ~ final_twa + Year + canvass2 + (1 | SIC.AI) + offset(log(offset_injuries))
## Data: .
##
## AIC BIC logLik deviance df.resid
## 33275.4 33308.8 -16631.7 33263.4 1931
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -40.395 -1.292 -0.159 1.365 18.760
##
## Random effects:
## Groups Name Variance Std.Dev.
## SIC.AI (Intercept) 0.1176 0.343
## Number of obs: 1937, groups: SIC.AI, 80
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 66.6767977 0.5168422 129.008 <2e-16 ***
## final_twa 0.0195833 0.0007610 25.733 <2e-16 ***
## Year -0.0331322 0.0002093 -158.276 <2e-16 ***
## canvass2Metal -0.6350735 0.2552916 -2.488 0.0129 *
## canvass2Nonmetal -0.4591329 0.2473421 -1.856 0.0634 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) fnl_tw Year cnvs2M
## final_twa -0.581
## Year -0.877 0.568
## canvass2Mtl -0.443 -0.007 -0.003
## cnvss2Nnmtl -0.459 -0.003 -0.002 0.932
## optimizer (Nelder_Mead) convergence code: 0 (OK)
## Model failed to converge with max|grad| = 0.0742931 (tol = 0.002, component 1)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
## Model is nearly unidentifiable: large eigenvalue ratio
## - Rescale variables?
## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: poisson ( log )
## Formula: n_fatal ~ final_twa + Year + (1 | SIC.AI) + offset(log(offset_fatal))
## Data: .
##
## AIC BIC logLik deviance df.resid
## 2300.5 2322.8 -1146.3 2292.5 1933
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.4468 -0.3749 -0.1797 -0.0733 9.6200
##
## Random effects:
## Groups Name Variance Std.Dev.
## SIC.AI (Intercept) 0.4518 0.6721
## Number of obs: 1937, groups: SIC.AI, 80
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 65.6324711 1.4964810 43.858 <2e-16 ***
## final_twa 0.0012618 0.0095534 0.132 0.895
## Year -0.0327034 0.0005867 -55.741 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) fnl_tw
## final_twa -0.637
## Year -0.845 0.134
## optimizer (Nelder_Mead) convergence code: 0 (OK)
## Model failed to converge with max|grad| = 0.218349 (tol = 0.002, component 1)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
## Model is nearly unidentifiable: large eigenvalue ratio
## - Rescale variables?
## Generalized linear mixed model fit by maximum likelihood (Laplace
## Approximation) [glmerMod]
## Family: poisson ( log )
## Formula:
## n_fatal ~ final_twa + Year + canvass2 + (1 | SIC.AI) + offset(log(offset_fatal))
## Data: .
##
## AIC BIC logLik deviance df.resid
## 2299.6 2333.0 -1143.8 2287.6 1931
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.4469 -0.3740 -0.1779 -0.0726 9.5934
##
## Random effects:
## Groups Name Variance Std.Dev.
## SIC.AI (Intercept) 0.3886 0.6234
## Number of obs: 1937, groups: SIC.AI, 80
##
## Fixed effects:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 65.8430208 1.5663506 42.036 <2e-16 ***
## final_twa 0.0032309 0.0096049 0.336 0.7366
## Year -0.0323792 0.0005868 -55.175 <2e-16 ***
## canvass2Metal -1.1009347 0.5082533 -2.166 0.0303 *
## canvass2Nonmetal -1.0504217 0.4720093 -2.225 0.0261 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) fnl_tw Year cnvs2M
## final_twa -0.607
## Year -0.815 0.134
## canvass2Mtl -0.247 -0.040 0.011
## cnvss2Nnmtl -0.290 -0.012 0.024 0.866
## optimizer (Nelder_Mead) convergence code: 0 (OK)
## Model failed to converge with max|grad| = 0.232036 (tol = 0.002, component 1)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
## Model is nearly unidentifiable: large eigenvalue ratio
## - Rescale variables?
