Factors Influencing Inadequate Finances Among Elderly Rural Women Submitted as Project Two for CHS 627 Spring 2000 1 Logistic Regression Logistic regression will be used to analyze the probability of a woman having inadequate (INADEQ$) finances, given the state of her health (NEWHLTH) and the number of years she was employed outside the home (YRSEMPLY). Because the logistic procedure makes none of the customary assumptions regarding equal variance and normal distribution, there is no need to review here the distribution of YRSEMPLY. Inadequate finances by Number of Years Employed If the logistic regression model uses only years of employment to predict inadequate finances, the results are not significant, as is shown below: Total number of cases: 228 (Unweighted) Number of selected cases: 228 Number of unselected cases: 0 Number of selected cases: 228 Number rejected because of missing data: 15 Number of cases included in the analysis: 213 Dependent Variable Encoding: Original Internal Value Value 0 0 1 1 _ Dependent Variable.. INADEQ$ Inadequate finances? Beginning Block Number 0. Initial Log Likelihood Function -2 Log Likelihood 262.06528 * Constant is included in the model. Beginning Block Number 1. Method: Enter Variable(s) Entered on Step Number 1.. YRSEMPLY # Yrs employed Estimation terminated at iteration number 3 because parameter estimates changed by less than .001 -2 Log Likelihood Goodness of Fit Cox & Snell - R^2 Nagelkerke - R^2 261.920 213.027 .001 .001 2 Chi-Square Model Block Step df Significance .145 .145 .145 1 1 1 .7029 .7029 .7029 The model is not significant. How well does the model fit the data? Ho: Model is a good fit. ---------- Hosmer and Lemeshow Goodness-of-Fit Test----------INADEQ$ = no, adequate fin INADEQ$ = yes, inadequate Group Observed Expected Observed Expected Total 1 2 3 4 5 6 7 8 9 10 14.000 15.000 13.000 14.000 15.000 17.000 16.000 15.000 4.000 25.000 15.043 14.158 13.379 14.016 14.645 16.661 14.491 13.691 4.766 27.152 7.000 5.000 6.000 6.000 6.000 7.000 5.000 5.000 3.000 15.000 5.957 5.842 5.621 5.984 6.355 7.339 6.509 6.309 2.234 12.848 21.000 20.000 19.000 20.000 21.000 24.000 21.000 20.000 7.000 40.000 The null hypothesis cannot be rejected, indicating that the Goodness-of-fit test 2.3344 8 .9690 model fits the data. There is not ----------------------------------------------------much difference between the expected and observed values. Chi-Square df Significance Classification Table for INADEQ$ The Cut Value is .50 Observed no, adequate fin n yes, inadequate y Predicted no, adequate finyes, inadequate Percent Correct n I y +---------------+---------------+ I 148 I 0 I 100.00% +---------------+---------------+ I 65 I 0 I .00% +---------------+---------------+ Overall 69.48% The classification table shows that this model correctly predicts only 69.48% of the cases. It fails to correctly predict any of the women with inadequate finances. ----------------- Variables in the Equation -----------------Variable B S.E. Wald df Sig R YRSEMPLY Constant -.0035 -.7482 .0091 .2447 .1450 9.3457 1 1 .7034 .0022 .0000 3 Given the significance level of .7034, the null hypothesis that the slope of the predictor (INADEQ$) is equal to zero cannot be rejected. The model is not significant; years of employment is not a significant predictor of inadequate finances. The equation for predicting inadequate finances using this model is: P(INADEQ$yes) = e-.7482-(.0035*YRSEMPLY) / 1 + e-.7482-(.0035*YRSEMPLY) Below is a plot of this equation. .33 .32 .31 .30 .29 .28 .27 .26 -20 0 20 40 60 80 # Yrs employed The relationship revealed by the plot would probably be better explained using simple linear regression, but that will be ignored for this project. The negative coefficients indicate an inverse relationship between the number of years a woman worked outside the home and the probability of her currently having inadequate finances; as the number of years employed increases, the probability of inadequate finances decreases. Variable Exp(B) YRSEMPLY .9965 95% CI for Exp(B) Lower Upper .9790 1.0144 4 The odds ratio is .9965; for every year a woman was employed, the probability that she has inadequate finances decreases by a measly 1.0035 (1/.9965). We know from above that the relationship is not significant; the 95% C.I. of (.99, 1.02) supports the finding. To illustrate, compared to woman who was employed for 10 years, a woman who was never employed has a 4% increased risk of having inadequate finances (e-.0035*(10-0), O.R.=1.04). When compared to a woman with a 20 year employment history, she is at 7% increased risk, and compared to the two women who were employed 70 years, she is 1.28 times more likely to have inadequate finances. Observed Groups and Predicted Probabilities 160 + + I I I I F I I R 120 + + E I I Q I I U I I E 80 + y + N I yy I C I yn I Y I nny I 40 + nny + I nnn I I nnn I I nnnn I Predicted --------------+--------------+--------------+--------------Prob: 0 .25 .5 .75 1 Group: nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnyyyyyyyyyyyyyyyyyyyyyyyyyyyyyy Predicted Probability is of Membership for yes, inadequate finances The Cut Value is .50 Symbols: n - no, adequate finances y - yes, inadequate finances Each Symbol Represents 10 Cases. This chart shows no real difference between the predicted probabilities of the groups, as might be expected, given the nonsignificance of the model. As with the classification table, no cases are predicted to have inadequate finances. 5 In summary, according to this model (which was not significant), unless a woman had been employed for a really long time, her chances of having inadequate finances are not much worse than a woman who was never employed. No one is predicted to have inadequate finances. Inadequate Finances by Health Status If we use the categorical variable NEWHLTH (1 = poor, 2 = fair, 3 = good) to predict INADEQ$ in a logistic regression model, we find it to be a significant predictor, as is shown below: Total number of cases: 228 (Unweighted) Number of selected cases: 228 Number of unselected cases: 0 Number of selected cases: 228 Number rejected because of missing data: 9 Number of cases included in the analysis: 219 Dependent Variable Encoding: Original Value 0 1 Internal Value 0 1 Dependent Variable.. INADEQ$ Inadequate finances? Beginning Block Number 0. Initial Log Likelihood Function -2 Log Likelihood 268.06499 * Constant is included in the model. Beginning Block Number 1. Method: Enter Variable(s) Entered on Step Number 1.. NEWHLTH Health recoded Estimation terminated at iteration number 3 because Log Likelihood decreased by less than .01 percent. -2 Log Likelihood Goodness of Fit Cox & Snell - R^2 Nagelkerke - R^2 Model Block Step 257.409 217.867 .047 .067 Chi-Square 10.656 10.656 10.656 df Significance 1 .0011 1 .0011 1 .0011 6 The model is significant (p = .0011). Health status is a significant predictor of financial inadequacy. ---------- Hosmer and Lemeshow Goodness-of-Fit Test----------INADEQ$ = no, adequate fin INADEQ$ = yes, inadequate Group Observed Expected Observed Expected Total 1 2 3 76.000 70.000 7.000 74.724 72.551 5.725 17.000 43.000 6.000 18.276 40.449 7.275 93.000 113.000 13.000 Goodness-of-fit test Chi-Square .8692 df Significance 1 .3512 >Warning # 18585 >The Hosmer-Lemeshow statistic is calculated from fewer than 6 groups. >Sensitivity to departures from model fit is substantially reduced. -------------------------------------------------------------- The predicted values are close to the observed ones, and the .3512 significance value of the Hosmer-Lemeshow statistic indicates that the model is a good fit for the data. Classification Table for INADEQ$ The Cut Value is .50 Observed no, adequate fin Predicted no, adequate finyes, inadequate Percent Correct n I y +---------------+---------------+ I 146 I 7 I 95.42% +---------------+---------------+ I 60 I 6 I 9.09% +---------------+---------------+ Overall 69.41% n yes, inadequate y However, the model correctly predicts only 69.41% of the cases and only 9.1% of the cases with inadequate finances. ----------------- Variables in the Equation -----------------Variable B S.E. Wald df Sig R NEWHLTH Constant -.8240 1.0637 .2592 .6054 10.1048 3.0872 1 1 .0015 .0789 -.1739 The results show that health status is a significant (p = .0015) predictor of inadequate finances and allow us to construct the prediction equation: 7 P(INADEQ$yes) = e 1.0637-(.8240*NEWHLTH) / 1 + e 1.0637-(.8240*NEWHLTH) Variable NEWHLTH Exp(B) 95% CI for Exp(B) Lower Upper .4387 .2639 .7291 The odds ratio is not helpful because of the categorical nature of this variable. To better understand the predictive relationship of health status on inadequate finances, it is necessary to create dummy variable to represent the different categories of health status in the logistic regression. Total number of cases: 228 (Unweighted) Number of selected cases: 228 Number of unselected cases: 0 Number of selected cases: 228 Number rejected because of missing data: 9 Number of cases included in the analysis: 219 Dependent Variable Encoding: Original Value 0 1 _ Internal Value 0 1 NEWHLTH poor fair good/excellent Value Freq 1 2 3 13 110 89 Parameter Coding (1) (2) .000 1.000 .000 .000 .000 1.000 “Poor” health status was designated as the indicator category for this analysis. Dependent Variable.. INADEQ$ Inadequate finances? Beginning Block Number 0. Initial Log Likelihood Function -2 Log Likelihood 268.06499 * Constant is included in the model. Beginning Block Number 1. Method: Enter Variable(s) Entered on Step Number 1.. NEWHLTH Health recoded Estimation terminated at iteration number 3 because Log Likelihood decreased by less than .01 percent. -2 Log Likelihood 256.545 8 Goodness of Fit Cox & Snell - R^2 Nagelkerke - R^2 218.996 .051 .073 Chi-Square Model Block Step df Significance 11.520 11.520 11.520 2 2 2 .0032 .0032 .0032 The model is significant at the .0032 level. ---------- Hosmer and Lemeshow Goodness-of-Fit Test----------INADEQ$ = no, adequate fin INADEQ$ = yes, inadequate Group Observed Expected Observed Expected Total 1 2 3 76.000 70.000 7.000 75.999 70.000 7.000 17.000 43.000 6.000 17.001 43.000 6.000 93.000 113.000 13.000 Chi-Square Goodness-of-fit test .0000 df Significance 1 1.000 >Warning # 18585 >The Hosmer-Lemeshow statistic is calculated from fewer than 6 groups. >Sensitivity to departures from model fit is substantially reduced. Despite the warning, the model appears to be an excellent fit for the data; observed and expected values are nearly identical. Based on the significance level of 1.000 on the Hosmer and Lemeshow Goodness-of-Fit test, the null hypothesis that the model fits the data cannot be rejected. Classification Table for INADEQ$ The Cut Value is .50 Observed no, adequate fin yes, inadequate n y Predicted no, adequate finyes, inadequate Percent Correct n I y +---------------+---------------+ I 153 I 0 I 100.00% +---------------+---------------+ I 66 I 0 I .00% +---------------+---------------+ Overall 69.86% 9 Using the criteria in the Classification Table, only 69.86% of cases are correctly predicted by the model, which is almost identical to that predicted by YRSEMPLY. This model also fails to correctly predict any cases with inadequate finances. ------------------ Variables in the Equation ------------------Variable NEWHLTH NEWHLTH(1) NEWHLTH(2) Constant B S.E. Wald df Sig R -.3331 -1.3433 -.1542 .5891 .6177 .5563 10.6984 .3198 4.7299 .0768 2 1 1 1 .0048 .5717 .0296 .7817 .1581 .0000 -.1009 The significance level of .0048 indicates that the variable NEWHLTH is a significant predictor of inadequate finances. Using the dummy variables NEWHLTH1 and NEWHLTH2, the prediction equation can be written as: P(INADEQ$yes) = e-.1542 - (.3331*NEWHLTH1) - (1.3433*NEWHLTH2) / 1 + e-.1542 - (.3331*NEWHLTH1) - (1.3433*NEWHLTH2) Variable NEWHLTH(1) NEWHLTH(2) Exp(B) .7167 .2610 95% CI for Exp(B) Lower Upper .2259 .0778 2.2739 .8757 The odds ratios show that women in poor health are 1.4 (1/.7167) times as likely as women fair health to have inadequate finances, but the 95% C.I. (.81, 1.99) means there is no difference in the probability of inadequate finances between the women in poor health and those whose health is fair. However, when compared to women in good health, those in poor health are 3.8 times more likely (95% C.I. (2.62, 5.04)) to have inadequate finances. Observed Groups and Predicted Probabilities F R E Q U E N 160 + I I I 120 + I I I 80 + I y n n + I I I + I I I + I y y y y n 10 C Y I n n I I n n I 40 + n n + I n n I I n n I I n n n I Predicted --------------+--------------+--------------+--------------Prob: 0 .25 .5 .75 1 Group: nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnyyyyyyyyyyyyyyyyyyyyyyyyyyyyyy Predicted Probability is of Membership for yes, inadequate finances The Cut Value is .50 Symbols: n - no, adequate finances y - yes, inadequate finances Each Symbol Represents 10 Cases. The plot shows that no cases are predicted to have inadequate finances. Inadequate Finances by Health Status and Years Employed Although years of employment was not a significant predictor by itself, we will add it to the model with health status to see if it achieves significance when combines with health status, or if it improves the health status model. Total number of cases: 228 (Unweighted) Number of selected cases: 228 Number of unselected cases: 0 Number of selected cases: 228 Number rejected because of missing data: 16 Number of cases included in the analysis: 212 Dependent Variable Encoding: Original Value 0 1 NEWHLTH poor fair good/excellent Internal Value 0 1 Value Freq 1 2 3 13 110 89 Parameter Coding (1) (2) .000 1.000 .000 .000 .000 1.000 Dependent Variable.. INADEQ$ Inadequate finances? Beginning Block Number 0. Initial Log Likelihood Function -2 Log Likelihood 261.33505 * Constant is included in the model. 11 Beginning Block Number 1. Method: Enter Variable(s) Entered on Step Number 1.. NEWHLTH Health recoded YRSEMPLY # Yrs employed Estimation terminated at iteration number 3 because Log Likelihood decreased by less than .01 percent. -2 Log Likelihood Goodness of Fit Cox & Snell - R^2 Nagelkerke - R^2 251.028 212.106 .047 .067 Chi-Square Model Block Step df Significance 10.307 10.307 10.307 3 3 3 .0161 .0161 .0161 The model is significant at the .0161 level. ---------- Hosmer and Lemeshow Goodness-of-Fit Test----------INADEQ$ = no, adequate fin INADEQ$ = yes, inadequate Group Observed Expected Observed Expected Total 1 2 3 4 5 6 7 8 9 10 16.000 15.000 18.000 23.000 14.000 11.000 13.000 15.000 15.000 7.000 17.055 15.393 16.987 22.565 13.121 11.176 13.604 13.549 16.550 7.000 5.000 4.000 3.000 5.000 7.000 7.000 9.000 7.000 12.000 6.000 3.945 3.607 4.013 5.435 7.879 6.824 8.396 8.451 10.450 6.000 21.000 19.000 21.000 28.000 21.000 18.000 22.000 22.000 27.000 13.000 Chi-Square Goodness-of-fit test 1.7741 df Significance 8 .9872 The model appears to be an excellent fit for the data, with not much difference between observed and expected values. Based on the significance level of .9872 on the Hosmer and Lemeshow Goodness-of-Fit test, the null hypothesis that the model fits the data cannot be rejected. Classification Table for INADEQ$ The Cut Value is .50 Predicted 12 Observed no, adequate fin n yes, inadequate y no, adequate finyes, inadequate Percent Correct n I y +---------------+---------------+ I 147 I 0 I 100.00% +---------------+---------------+ I 65 I 0 I .00% +---------------+---------------+ Overall 69.34% As in the previous models, only about 69% of cases are correctly predicted by the model. This model also fails to correctly predict any cases with inadequate finances. ------------------ Variables in the Equation ------------------Variable NEWHLTH NEWHLTH(1) NEWHLTH(2) YRSEMPLY Constant B S.E. Wald df Sig R -.3218 -1.2811 -.0011 -.1374 .5922 .6222 .0092 .5749 9.4763 .2953 4.2393 .0134 .0571 2 1 1 1 1 .0088 .5869 .0395 .9080 .8111 .1448 .0000 -.0926 .0000 When controlled for YRSEMPLY, NEWHLTH is significant in the model (.0088), but controlling for health status does not change the lack of significance (.9080) of YRSEMPLY as a predictor. The negative values of the coefficients are also as expected: As the number of years a woman was employed outside the home decreases, the probability of her having inadequate finances increases; similarly, as the state of heath improves, the probability of having inadequate finances decreases. The small coefficient for YRSEMPLY (-.0011) demonstrates how little it adds to the equation. The equation for predicting inadequate finances using this model is: P(INADEQ$yes) = e-.1374 - (.3218*NEWHLTH1 - 1.2811*NEWHLTH2-.0011*YRSEMPLY) / 1 + e-.1374 - (.3218*NEWHLTH1 - 1.2811*NEWHLTH2-.0011*YRSEMPLY) Variable NEWHLTH(1) NEWHLTH(2) YRSEMPLY Exp(B) .7249 .2777 .9989 95% CI for Exp(B) Lower Upper .2271 .0820 .9810 2.3137 .9403 1.0172 13 The odds ratios are also very close to those of the model with health status alone. Women in poor health are equally as likely (95% C.I. (.23, 2.3)) as women fair health to have inadequate finances. They are 3.6 (95% C.I. (2.3, 4.8)) times more likely than women in good health to have inadequate finances. Observed Groups and Predicted Probabilities 160 + + I I I I F I I R 120 + + E I I Q I I U I y I E 80 + y + N I n I C I n y I Y I n yy I 40 + n ny + I n nn I I n nn I I n nn n I Predicted --------------+--------------+--------------+--------------Prob: 0 .25 .5 .75 1 Group: nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnyyyyyyyyyyyyyyyyyyyyyyyyyyyyyy Predicted Probability is of Membership for yes, inadequate finances The Cut Value is .50 Symbols: n - no, adequate finances y - yes, inadequate finances Each Symbol Represents 10 Cases. This plot also shows that the failure of the model to correctly predict cases of inadequate finances. Multicollinearity Because relationships among predictor variables can influence outcome, one-way analysis of variance is used to investigate multicollinearity between the two predictor variables. Testing the assumptions of a normal distribution shows that values for YRSEMPLY ranged from a minimum of zero (which was also the mode) to a maximum of 70 years. The median number of 14 years employed was 20, and the mean was 21.63, with a variance of 267.05. Skewnness of .37 and kurtosis of -.412 indicate a relatively normal distribution, and the histogram is reasonably normal, although a large number (n = 41) of women were never employed. Statistics YRSEMPLY # Yrs employed N Valid Missing Mean Median Mode Std. Deviation Variance Skewness Std. Error of Skewness Kurtosis Std. Error of Kurtosis Range Minimum Maximum Frequency 221 7 21.63 20.00 0 16.34 267.05 .370 .164 -.412 .326 70 0 70 # Yrs employed 50 40 30 20 10 Std. Dev = 16.34 Mean = 21.6 N = 221.00 0 0.0 10.0 5.0 20.0 15.0 30.0 25.0 40.0 35.0 50.0 45.0 55.0 60.0 70.0 65.0 # Yrs employed Test of Homogeneity of Variances YRSEMPLY # Yrs employed Levene Statistic .622 df1 2 df2 216 Sig. .538 We cannot reject the null hypothesis of equal variance. Based on the significance level of .538, the null hypothesis that the mean number of years employed is equal across the 3 categories of health status cannot be rejected. ANOVA YRSEMPLY # Yrs employed Between Groups Within Groups Total Sum of Squares 836.344 57307.136 58143.479 df 2 216 218 Mean Square 418.172 265.311 15 F 1.576 Sig. .209 Descriptives YRSEMPLY # Yrs employed N 1 poor 2 fair 3 good/excellent Total 13 116 90 219 Mean 15.69 21.06 23.59 21.78 Std. Deviation 17.08 16.81 15.48 16.33 Std. Error 4.74 1.56 1.63 1.10 95% Confidence Interval for Mean Lower Upper Bound Bound 5.37 26.01 17.97 24.15 20.35 26.83 19.61 23.96 Minimum 0 0 0 0 Although the mean number of years employed increases as the health status improves, the differences are not significant, perhaps a consequence of the small (n = 13) number of women in the “poor” health category. Log-Linear Analysis The number of years a woman was employed has not been shown to have a significant effect on her current financial status, perhaps because so many of the women were never employed. The following analysis will substitute the categorical variable EVEREMPL, which has a value of “1” if a woman was ever employed and “2” if she was not. The Log-Linear method will be used to examine the relationships, including possible interactions, among inadequate finances, health status, and whether or not a woman was ever employed. * * * * * * * * DATA H I E R A R C H I C A L L O G L I N E A R * * * * * * * * Information 219 unweighted cases accepted. 0 cases rejected because of out-of-range factor values. 9 cases rejected because of missing data. 219 weighted cases will be used in the analysis. FACTOR Information Factor Level NEWHLTH 3 INADEQ$ 2 EVEREMPL 2 Label Health recoded Inadequate finances? Ever employed? 16 Maximum 45 70 70 70 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - * * * * * * * * H I E R A R C H I C A L L O G L I N E A R * * * * * * * * DESIGN 1 has generating class NEWHLTH*INADEQ$*EVEREMPL Note: For saturated models .500 has been added to all observed cells. This value may be changed by using the CRITERIA = DELTA subcommand. The Iterative Proportional Fit algorithm converged at iteration 1. The maximum difference between observed and fitted marginal totals is and the convergence criterion is .250 .000 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Observed, Expected Frequencies and Residuals. Factor Code NEWHLTH INADEQ$ EVEREMPL EVEREMPL INADEQ$ EVEREMPL EVEREMPL poor no, adeq yes no yes, ina yes no NEWHLTH INADEQ$ EVEREMPL EVEREMPL INADEQ$ EVEREMPL EVEREMPL fair no, adeq yes no yes, ina yes no NEWHLTH INADEQ$ EVEREMPL EVEREMPL INADEQ$ EVEREMPL EVEREMPL good/exc no, adeq yes no yes, ina yes no OBS count EXP count Residual Std Resid 5.5 2.5 5.5 2.5 .00 .00 .00 .00 4.5 2.5 4.5 2.5 .00 .00 .00 .00 58.5 12.5 58.5 12.5 .00 .00 .00 .00 33.5 10.5 33.5 10.5 .00 .00 .00 .00 65.5 11.5 65.5 11.5 .00 .00 .00 .00 14.5 3.5 14.5 3.5 .00 .00 .00 .00 The Frequency Table shows that some of the cells have a small n, but none are empty. - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Goodness-of-fit test statistics Likelihood ratio chi square = .00000 DF = 0 P = 1.000 Pearson chi square = .00000 DF = 0 P = 1.000 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Tests that K-way and higher order effects are zero. K DF L.R. Chisq Prob Pearson Chisq 17 Prob Iteration 3 2 1 2 .037 .9818 .037 7 14.252 .0469 14.756 11 244.085 .0000 290.425 - - - - - - - - - - - - - - - - - - - - - - - - - - - - .9819 .0393 .0000 - - - - - - - - 3 2 0 - - - - Tests that K-way effects are zero. K DF L.R. Chisq Prob Pearson Chisq Prob Iteration 1 2 3 4 5 2 229.834 14.215 .037 .0000 .0143 .9818 275.669 14.719 .037 .0000 .0116 .9819 0 0 0 The interaction of inadequate finances, health status, and employed is not significant (.9819). * * * * * * * * H I E R A R C H I C A L L O G L I N E A R * * * * * * * * Tests of PARTIAL associations. Effect Name DF NEWHLTH*INADEQ$ NEWHLTH*EVEREMPL INADEQ$*EVEREMPL NEWHLTH INADEQ$ EVEREMPL - - - - - - - - - - - - - - - - - - - - - - - - - Partial Chisq 2 2 1 2 1 1 - - - - - Prob Iter 11.027 .0040 2 1.473 .4788 2 .729 .3933 2 98.921 .0000 2 35.534 .0000 2 95.380 .0000 2 - - - - - - - - - - Although the 3-way interaction was not significant, we look at the 2-way interactions and find that health status has a significant effect of inadequate finances (p = .004), but employment does not (p = .3933). These 2-way interactions could also have been examined with the results that are summarized below. Chi-Square 12.249 12.249 12.249 Model Block Step df Significance 3 .0066 3 .0066 3 .0066 The model is significant. ---------- Hosmer and Lemeshow Goodness-of-Fit Test----------INADEQ$ Group = no, adequate fin INADEQ$ Observed Expected = yes, inadequate Observed Expected 18 Total 1 2 3 4 65.000 11.000 58.000 19.000 65.175 10.824 57.754 19.246 14.000 3.000 33.000 16.000 Chi-Square Goodness-of-fit test 13.825 3.176 33.246 15.754 79.000 14.000 91.000 35.000 df Significance .0253 2 .9875 >Warning # 18585 >The Hosmer-Lemeshow statistic is calculated from fewer than 6 groups. >Sensitivity to departures from model fit is substantially reduced. -------------------------------------------------------------- The model is a good fit for the data. Classification Table for INADEQ$ The Cut Value is .50 Observed no, adequate fin n yes, inadequate y Predicted no, adequate finyes, inadequate Percent Correct n I y +---------------+---------------+ I 151 I 2 I 98.69% +---------------+---------------+ I 64 I 2 I 3.03% +---------------+---------------+ Overall 69.86% The model correctly classified 70% of cases but only 1% of those with inadequate finances. Variable NEWHLTH NEWHLTH(1) NEWHLTH(2) EVEREMPL Constant B S.E. Wald df Sig R -.2976 -1.2960 .3247 -.5794 .5921 .6213 .3772 .7455 10.2634 .2526 4.3515 .7407 .6039 2 1 1 1 1 .0059 .6153 .0370 .3894 .4371 .1529 .0000 -.0937 .0000 When controlled for whether or not a women was ever employed, health status was shown to be a significant predicator of inadequate finances (p = .0059). EVEREMPL, controlled for NEWHLTH was not. Variable Exp(B) NEWHLTH(1) NEWHLTH(2) EVEREMPL .7426 .2736 1.3836 95% CI for Exp(B) Lower Upper .2327 .0810 .6605 2.3700 .9247 2.8980 19 The 95% C.I. for the odds ratio of comparing ever-employed to never-employed includes 1, so the O.R. value of 1.4 cannot be interpreted as meaning women who have never been employed are 1.4 times more likely to have inadequate finances when health status is controlled. When controlled for whether or not they were ever employed, women in poor health are 1.35 (O.R. = 1/.7426) times more likely than women in fair health and 3.65 times more likely than women in good health to have inadequate finances. These figures are not much different from those in the logistic regression with health status alone. When the odds ratios of inadequate to adequate finances are calculated, controlling NEWHLTH for EVEREMPL and vice versa, the following results are obtained: Ever employed? Health Status yes no O.R. 95% C.I. O.R. 95% C.I. Poor – Fair 1.4 -.54, 3.35 1.2 -1.36, 3.76 Poor - Good 3.7 -1.62, 9.05 3.7 -4.91, 12.24 Fair - Good 2.6 .74, 4.54 3.1 -1.61, 7.72 Employment appears to have the greatest effect on inadequate finances when the when those in fair health are compared to those in good health. Among women who never were employed, the odds for women in fair health having inadequate finances are 3.1 times those for women in good health. but the odds are only 2.6 for those women who were employed. Women in poor health are 3.7 times as likely as women in good health to have inadequate finances, regardless of whether they were employed or not. However, if the confidence intervals are correct, none of these odds ratios can be considered with 95% confidence. 20 The graph below, comparing health status of the employed and never-employed women who have inadequate finances, shows higher percentages of inadequate finances for women who were never employed, regardless of health status. Although the lines are not parallel, they are nearly so on both segments, as might be expected given the lack of significance of the interaction. Inadequate Finances by Health Status and Employed 60% 50% employed 40% never employed 30% 20% 10% 0% poor fair good Health Status Inadequate Finances by Employment 40% 37.5% 35% 30% 25% 28.5% 20% 15% 10% 5% 0% yes no Ever Employed? 21 The graph of EVEREMPL and INADEQ$ shows that a higher percentage of women who were never employed have inadequate finances, although this 2-way interaction was not significant. Inadequate Finances by Health Status 50% 45% 40% 46.2% 38.1% 35% 30% 25% 20% 18.3% 15% 10% 5% 0% poor fair good Health Status The graph of the significant effect of health status on inadequate finances shows that as the state of health improves, the percentage of women with inadequate finances decreases. 22 MANOVA (2 outcomes, 1 predictor) As discussed in the previous paper, this study used measures of personal control score and autonomy as outcome variables. These measures were described in more detail previously, but to summarize them briefly: Personal control was measured using an 8-item instrument that was a modified version of an existing instrument. Items were measured using a four-point Likert-type scale with responses ranging from “strongly disagree” to “strongly agree,” with values of one to four, respectively. Items were summed to produce an overall score ranging from 8 to 32; higher scores indicate greater sense of personal control. Autonomy was measured using a existing 14-item inventory with a reported internal consistency of .83. Items were measured using a six-point Likert-type scale with responses ranging from “strongly disagree” to “strongly agree,” with value of one to six, respectively. Items were summed to produce an overall score that could range from 14 to 84; higher scores indicate greater sense of autonomy. Personal control scores were obtained for 227 women and ranged from a low score of 12 to a high score of 32. The median score was 23, and the mean score was 23.20, with a variance of 14.86. Autonomy scores, obtained for 226 women, ranged from 25 to 84, with a median score of 57, a mean of 58.17, and a variance of 157.46. Skewness and kurtosis statistics, as well as histograms of the frequency distributions for both variables indicated reasonably normal distributions of scores. Statistics N Mean Median Valid Missing PCSCORE 227 1 23.1982 23.0000 AUTSCORE 226 2 58.1726 57.0000 23 Mode Std. Deviation Variance Skewness Std. Error of Skewness Kurtosis Std. Error of Kurtosis Range Minimum Maximum 24.00 3.8547 14.8588 -.095 .162 .186 .322 20.00 12.00 32.00 49.00 12.5483 157.4590 .183 .162 -.502 .322 59.00 25.00 84.00 PCSCORE AUTSCORE 70 50 60 40 50 30 40 20 20 Std. Dev = 3.85 10 Mean = 23.2 N = 227.00 0 12.0 16.0 14.0 20.0 18.0 24.0 22.0 28.0 26.0 Frequency Frequency 30 32.0 30.0 PCSCORE 10 Std. Dev = 12.55 Mean = 58.2 N = 226.00 0 25.0 35.0 30.0 45.0 40.0 55.0 50.0 65.0 60.0 75.0 70.0 85.0 80.0 A UTSCORE To examine the effect of inadequate finances on these scores simultaneously requires a multivariate analysis of variance (MANOVA). b j l u N b I 0 N n I n 2 a f f i 1 y 6 i n f 24 e t I d N e ia N f a i P 0 2 0 2 1 2 2 6 f i T 5 9 8 A 0 6 2 2 1 3 7 6 f i T 7 6 8 a o B F d d S T o a D The significance level of .749 means that the null hypothesis of equal covariance structures in both groups cannot be rejected, and the assumption of equal variance has not been violated. In other words, although differences in the covariances are seen in the following Correlations table, they are not significant (PCSCORE covariances: poor = 13.590, fair = 12.931, good = 15.615; AUTSCORE covariances: poor = 87.526, fair = 170.491, good = 138.765). The table also shows moderate correlations between PCSCORE and AUTSCORE for all levels of health; for those in fair and good health, the correlations are significant at p < .001. 25 r e l a t i o A U N T E S W C C S H R C e E O a l . P P C e a S 0 0 0 0 0 0 * * S ig . . . S u m 0 0 0 0 0 0 C r o C o v 0 0 0 0 0 0 N 2 2 A P U e a T 0 0 0 0 0 0 * * S ig . . . S u m 0 5 0 0 0 0 C r o C o v 0 5 0 0 0 0 N 2 2 1 P P C e p a S o 0 4 0 9 0 7 S ig . 0 8 4 . S u m 0 8 7 4 7 6 C r o C o v 5 1 9 5 0 4 N 1 1 3 3 A P U e a T 4 0 9 0 7 0 S ig . 0 8 4 . S u m 8 3 4 0 6 8 C r o C o v 1 5 5 2 4 6 N 1 1 3 3 2 P P C e f a a S 0 5 0 5 0 2 * * S i g . 0 0 0 . S u m 8 0 4 5 9 0 C r o C o v 9 9 3 3 1 3 N 1 1 1 1 9 9 A P U e a T 5 0 5 0 2 0 * * S ig . 0 0 0 . S u m 0 9 5 8 0 3 C r o C o v 9 4 3 9 3 1 N 1 1 1 1 9 9 3 P P C e g a S o 0 5 0 8 0 4 * * S ig . 0 0 0 . S u m 5 6 8 7 1 4 C r o C o v 6 4 1 0 5 3 N 9 9 3 2 A P U e a T 5 0 8 0 4 0 * * S ig . 0 0 0 . S u m 6 6 7 0 4 9 C r o C o v 4 7 0 6 3 5 N 9 9 2 2 * *o . C r r 26 Continuing discussion of the MANOVA below, using an alpha level of .06 allows us to reject the null hypothesis of no difference between the mean vectors of the two groups: Ho: pc1 = pc2 aut1 = aut1 and accept the model as significant. i b a o t h S a d F o E l i u g f a I P n 0 8 0 0 0 a W 0 8 0 0 0 a H 4 8 0 0 0 a R 4 8 0 0 0 a I P N 6 7 0 0 7 a W 4 7 0 0 7 a H 7 7 0 0 7 a R 7 7 0 0 7 a E b D We investigate the significance by first examining the univariate results. a a l F f f i g 1 2 P 7 1 6 0 A 1 1 6 8 T v a D There is no problem with variance at the univariate level, even before applying the Bonferroni correction. 27 n - p e M u m e a q u S d S u D a F i f a g o a C P 7 3 2 9 1 9 4 0 b A 4 9 7 0 1 0 7 3 I P n 6 9 0 9 1 9 6 0 A 7 9 0 3 1 3 3 0 I P N 7 3 2 9 1 9 4 0 A 4 9 7 0 1 0 7 3 E P r 8 7 6 8 A 1 9 6 6 T P o 0 8 A 0 8 C P 6 7 A 9 7 a . R b . R The means from the Descriptive Statistics table showed that women with inadequate finances had lower scores in both personal control and autonomy than did women with adequate finances. The Between-Subjects Effects table shows that when personal control scores and autonomy scores are considered simultaneously, those differences are significantly lower on personal control scores (p = .02), but there is no significant difference in autonomy scores (p = .373). Even after controlling for using the Bonferroni-corrected value of .025 (.05/2), personal control scores for women with inadequate finances are significantly lower. The following pairwise comparisons show the same thing and are not necessary since there are only two groups, they are included for example only. m d e p w p o o e D u u I E a n P 0 8 9 0 7 1 8 9 5 2 f i A 0 2 0 2 1 1 7 7 8 7 f i 28 C o m d e a f f e e a e o p r w e p a S o o . I D ( ( u i u E J I g J P 0 1 1 2 0 1 0 4 6 * f i 1 0 2 1 0 1 0 6 4 * f i n A 0 1 U 9 0 4 3 3 8 7 f i 1 0 0 9 4 3 3 7 8 f i n B i *. T h a . A d a o t h S a d F o l i u g f a P 6 7 0 0 7 a W 4 7 0 0 7 a H 7 7 0 0 7 a R 7 7 0 0 7 E o m a E This table also shows, at = .06, the significance of the multivariate model. a t m e a S u d u F D a i g f P C 9 1 9 4 0 E 7 6 8 A C 0 1 0 7 3 E 9 6 6 T p This table also shows again the significance levels of the univariate tests. 29 Estimated Marginal Means of PCSCORE 23.8 23.6 Estimated Marginal Means 23.4 23.2 23.0 22.8 22.6 22.4 22.2 no, adequate finance yes, inadequate fina Inadequate finances? This plot shows that the lower personal control scores for women with inadequate finances when personal control and autonomy scores are considered simultaneously. Estimated Marginal Means of AUTSCORE 59.0 Estimated Marginal Means 58.5 58.0 57.5 57.0 no, adequate finance yes, inadequate fina Inadequate finances? Similarly, this plot shows that the lower autonomy scores for women with inadequate finances when personal control and autonomy scores are considered simultaneously. 30 Discriminant Function Analysis Discriminant Function Analysis analyzes the relationships among these variables (INADEQ$, PCSCORE, AUTSCORE) in a different way and allows the construction of an equation, using PCSCORE and AUTSCORE, that best discriminates between those with inadequate and adequate finances. o c r N U c V 8 6 E M 8 5 g A 2 9 d B o 0 0 a d T 0 4 T 8 0 S t ( l I e g f i g 0 P 2 0 A 2 0 1 P 6 0 f A 6 0 T P 8 0 A 8 0 a G C C C P 8 4 A 4 6 C P 0 7 A 7 0 a T 31 Because we learned form the MANOVA that the Box’s test was not significant, we can pool the within-groups covariance matrices. The variables PCSCORE and AUTSCORE are moderately correlated at .567. Box's Test of Equality of Covariance Matrices e r o I a m f n 0 2 7 1 2 7 f P 2 8 T t h R 2 6 3 8 9 e B F A d d S T The Box’s Test is not significant, allowing us to pool the variance. Summary of Canonical Discriminant Functions v o u o n e n % a F l v a 1 0 0 2 a F a 32 L l k m d i q T g 1 7 i c 1 P A The Standardized Canonical Discriminant Function Coefficients allows comparison of the amounts each variable contributes to the differentiation between the INADEQ$ groups. The PCSCORE coefficient (1.128) is four times larger than that of AUTSCORE (-.269), reflecting its greater importance in differentiating between the groups; it is also in a different direction from the AUTSCORE coefficient. These coefficients can be used to calculate a new variable, x, that uses standardized scores for personal control and autonomy to give the best separation of women with and without adequate finances. The equation for calculating the variable is: x = (1.128*PCSCOREstd) – (.269*AUTSCOREstd) r c I N 1 f i 0 8 1 8 f i U g 33 r e c 1 P A P v V Classification Statistics c P E M g A d U t i d I N g f i i o g 0 0 2 0 1 0 6 0 f T 8 0 c I n c y n e q q u n n c c P A ( C F 34 i a o e d b e y n e e q I q N u a o n n f i t O C 0 6 6 2 1 9 7 6 f i U 5 3 8 % 0 6 4 0 1 9 1 0 f i U 5 5 0 a 5 Only 56.4% of cases were correctly classified, which is little better than chance. MANOVA (2 outcomes, 2 predictors) Perhaps ethnicity, combined with inadequate finances, has an effect on the outcomes. NEWETH is a categorical variable with the value of 1 for Caucasian and 2 for African American. Adding NEWETH into the analysis gives the following results: b j N I 0 n 2 I f f 1 y 2 f N 1 3 C E 2 1 A The descriptive statistics that follow reveal higher mean personal control scores and autonomy scores for African American women when compared to Caucasian women, regardless of financial status. All cells have a reasonable number of cases. 35 v e S I N N td M v e f i r N i a e n a P 0 1 C 0 2 2 8 7 2 5 2 0 1 5 T 2 5 2 0 2 1 1 0 4 1 2 6 f i n 2 9 1 0 1 6 T 9 6 7 9 2 T 1 o 7 7 2 4 3 2 4 4 9 9 1 T 4 1 1 4 4 A 0 1 U 9 2 2 7 7 2 2 2 0 0 5 T 2 5 6 2 2 1 1 6 4 3 9 6 f i n 2 0 1 5 1 6 T 6 6 9 5 2 T 1 o 4 7 3 0 3 2 0 4 4 6 1 T 6 1 9 2 4 a o B F d d S T o a D N Box’s Test of the null hypothesis of equal covariance matrices cannot be rejected. 36 r b ia o th r a S r E d F o l i u f g f a I P n t 0 0 9 5 0 0 0 a W 0 0 1 5 0 0 0 a H 0 0 1 5 0 0 0 a R 0 0 1 5 0 0 0 a I P N 0 5 8 6 0 0 1 a W 0 5 2 6 0 0 1 a H 0 5 8 6 0 0 1 a R 0 5 8 6 0 0 1 a N P E 0 8 4 4 0 0 3 a W 0 8 6 4 0 0 3 a H 0 8 4 4 0 0 3 a R 0 8 4 4 0 0 3 a I P N 0 7 4 9 0 0 8 a W 0 7 6 9 0 0 8 a H 0 7 4 9 0 0 8 a R 0 7 4 9 0 0 8 a . E b . D All tests have the identical significance value of .678, indicating that the interaction of inadequate finances and ethnicity does not have a significant effect on personal control and autonomy scores when they are considered simultaneously. Nonetheless, let’s consider what a significant interaction would have meant. Had the interaction been significant for PSCORE, it would have indicated that finances and ethnicity, when considered together, have a significant effect on PSCORE when it is considered simultaneously with AUTSCORE. Similarly, a significant interaction for AUTSCORE would have indicated that finances and ethnicity, when considered together, have a significant effect on AUTSCORE when it is considered simultaneously with PCSCORE. This is discussed in more detail later. Since the interaction is not significant, we look at the significance values of the main effects and find that neither achives significance. 37 a a F f f i g 1 2 P 6 3 0 0 A 8 3 0 4 T v a D N The null hypothesis of equal error variance cannot be rejected. n -S p e M u m e a q u S d S u D a F i f a g o a C P 9 9 3 9 3 0 5 1 b A 5 0 6 6 3 2 0 1 I P n t 6 3 0 2 1 2 2 0 A 2 0 0 4 1 4 3 0 I P N 5 2 5 4 1 4 4 2 A 6 8 0 5 1 5 2 0 N P E 5 3 8 1 1 1 0 9 A 1 1 2 8 1 8 4 6 I P N 9 7 9 7 1 7 0 2 A 8 1 9 5 1 5 6 8 E P r 9 1 0 3 A 6 2 0 9 To P 0 4 A 0 4 C P 0 3 A 8 3 a . R b . R When INADEQ$ is controlled for ETHNICITY, autonomy scores are not significant (p = .300), but personal control scores are significant in the table at .052, indicating a significant effect if is set at .10. Even after applying the Bonferroni correction, rounding, and comparing .05 to the corrected value of .05 (.10/2), personal control scores are significant. Personal control scores are not significantly affected by the ethnicity variable (p = .289) when it is 38 controlled for financial status, but autonomy scores are significant in the table (p = .026) and remain so after applying the Bonferroni correction with = .10 (.026 < .05). As discussed previously, the interaction of ethnicity and inadequate finances was not significant, which indicates that the combined effect of NEWETH and INADEQ$ is not significant when personal control and autonomy scores are considered simultaneously. We look at marginal means to learn more about these differences. Estimated Marginal Means 1. Inadequate finances? m d e p w p o o e D u u I E a n P 0 4 4 9 9 1 1 9 9 2 f i A 0 0 4 1 9 1 1 6 0 1 f i o m d e a f f a e e e o p r w p e a S o . o I D ( ( i u E u J I g J P 0 1 9 4 7 2 2 8 f i 1 0 0 4 7 2 8 2 f i n A 0 1 7 9 9 0 5 3 f i 1 0 9 9 9 0 3 5 f i n B a . A 39 Estimated Marginal Means of PCSCORE 24.2 24.0 23.8 Estimated Marginal Means 23.6 23.4 23.2 23.0 22.8 22.6 22.4 no, adequate finance yes, inadequate fina Inadequate finances? Estimated Marginal Means of AUTSCORE 61.5 61.0 Estimated Marginal Means 60.5 60.0 59.5 59.0 58.5 no, adequate finance yes, inadequate fina Inadequate finances? When adjusted for ethnicity, women with inadequate finances had significantly lower personal control scores than did women with adequate finances. As discussed above, although their autonomy scores were also lower, the difference was not significant. Pairwise comparisons (following) are not really necessary, since INADEQ$ has only two categories. 2. Ethnicity recoded 40 m e p w p e u D u E E a P 1 7 5 6 8 2 7 5 4 1 A 1 2 5 3 1 2 9 1 3 4 o d e a f f e e a e o p r w e p a S o o . I D ( ( i u u E J I g J P 1 2 0 7 9 5 4 2 1 0 7 9 4 5 A 1 2 6 9 6 0 3 * 2 1 6 9 6 3 0 * B * . T a A Estimated Marginal Means of PCSCORE 23.8 Estimated Marginal Means 23.6 23.4 23.2 23.0 22.8 Cauc asian Afric an-Americ an Ethnicity recoded 41 Estimated Marginal Means of AUTSCORE 63 62 Estimated Marginal Means 61 60 59 58 57 Cauc asian Afric an-Americ an Ethnicity recoded When adjusted for financial situation, both personal control and autonomy scores of African-American women were higher than those of Caucasians. As discussed above, the difference was significant for autonomy but not for personal control. Pairwise comparisons are not really necessary, since ethnicity has only two categories. e s d e o p w p o o . e D I u E u E n a P 0 1 8 5 7 0 2 0 6 9 1 1 1 6 7 7 5 f i 2 5 5 2 8 A 0 1 0 8 5 5 2 0 5 1 9 1 1 4 5 7 2 f i 2 7 4 8 7 interaction Profile Plots PCSCORE 42 Estimated Marginal Means of PCSCORE 25.0 24.5 Estimated Marginal Means 24.0 23.5 Inadequate finances? 23.0 no, adequate finance s 22.5 yes, inadequate fina 22.0 nces Cauc asian Afric an-Americ an Ethnicity recoded The lines in both graphs (above and below) are almost parallel, as is expected, given the lack of a significant interaction. African-American women have higher personal control scores than do Caucasian women, more so for women with adequate finances than for those without. Estimated Marginal Means of PCSCORE 25.0 24.5 Estimated Marginal Means 24.0 23.5 23.0 Ethnicity recoded 22.5 Cauc asian 22.0 Afric an-Americ an no, adequate finance yes, inadequate fina Inadequate finances? Seen another way, women with inadequate finances have lower personal control scores, but this difference is greater for African-Americans. 43 AUTSCORE Estimated Marginal Means of AUTSCORE 66 Estimated Marginal Means 64 62 Inadequate finances? 60 no, adequate finance s 58 yes, inadequate fina 56 nces Cauc asian Afric an-Americ an Ethnicity recoded This graph shows more evidence of an interaction, although it is not significant. For both ethnic groups, the scores of women with inadequate finances are lower than those of women whose financial situation is adequate, but this difference is much smaller for Caucasian women than for are African-Americans. Estimated Marginal Means of AUTSCORE 66 Estimated Marginal Means 64 62 60 Ethnicity recoded 58 Cauc asian 56 Afric an-Americ an no, adequate finance yes, inadequate fina Inadequate finances? 44 This graph also shows that, for Caucasian women, financial status has little effect on autonomy scores, but the scores for African-American women are lower for those whose financial situation is inadequate than are those of women with adequate finances. The table below shows not much difference in the covariances (excluding the missing values), which helps understand why the Box’s test was not significant. The correlations of personal control and autonomy scores from the table on the following page are summarized below: Adequate finances Caucasian n r p 127 .529 .000 26 .536 .006 46 .671 .000 16 .5786 .019 African-American Inadequate finances Caucasian African-American Personal control and autonomy scores are moderately and significantly correlated within all groups. 45 o r r e l a t i o n s I N N A E D W E Q E T $ H A P C U S T C S C O O R R E f i n r a e n c c o e d s e ? d . 1 P P C C e a S a u C r c s O a o 1 . 0 . 8 0 9 0 3 S i g . ( 2 . 1 0 7 . Su m o f 1 2 2 6 . 7 . 5 5 0 0 0 C r o s s C o v a r i 4 8 . 2 . 8 5 3 0 3 N 4 4 A P U eT aS r s C o O 1 . 8 . 0 9 0 3 0 S ig . ( 2 . 1 0 7 . Su m o f 2 6 6 9 . 5 . 0 0 0 0 0 C r o s s C o v a r i 2 8 3 . 8 . 0 3 0 3 0 N 4 4 2 P P A C e f r S a ic C ra s O n o 1 . 0 . 2 0 9 0 6 S i g . ( 2 . 7 0 4 . Su m o f 1 1 8 7 . 7 . 5 5 0 0 0 C r o s s C o v a r i 6 5 . 2 . 8 5 3 0 3 N 4 4 A P U eT aS r s C o O 1 . 2 . 0 9 0 6 0 S ig . ( 2 . 7 0 4 . Su m o f 1 1 8 7 7 . 5 . 0 0 0 0 0 C r o s s C o v a r i 6 5 2 . 8 . 3 3 3 3 3 N 4 4 0 1 n P P o C C e , a a S a u C d rc s e O a o q 1 . 0 . 5 0 2 0 9 * * S i g . ( 2 . 0 0 0 . Su m o f 1 5 1 3 0 . 7 . 1 3 1 2 0 C r o s s C o v a r i 1 2 3 4 . 9 . 6 1 8 9 3 N 1 1 2 2 7 7 A P U e T a S r s C o O 1 . 5 . 0 2 0 9 0 * * S ig . ( 2 . 0 0 0 . Su m o f 6 1 8 0 6 . 1 . 4 1 2 0 5 C r o s s C o v a r i 1 2 5 4 6 . 6 . 2 8 4 3 1 N 1 1 2 2 7 7 2 P P A C e f r S a ic C ra s O n o 1 . 0 . 5 0 3 0 6 * * S i g . ( 2 . 0 0 6 . Su m o f 5 7 5 0 7 . 0 . 2 0 0 0 0 C r o s s C o v a r i 1 2 8 3 . 8 . 2 0 1 0 7 N 2 2 6 5 A P U e T a S r s C o O 1 . 5 . 0 3 0 6 0 * * S ig . ( 2 . 0 0 6 . Su m o f 9 5 4 7 7 . 2 . 4 0 4 0 0 C r o s s C o v a r i 1 2 2 3 2 . 2 . 8 1 1 7 0 N 2 2 5 5 1 1 y P P e C C e s a S a , u C i r n c s O a a o d 1 . 0 . 6 0 7 0 1 * * f in a n c s S i ge . ( 2 . 0 0 0 . Su m o f 3 1 9 6 5 . 1 . 4 0 3 9 5 C r o s s C o v a r i 1 3 3 1 . 6 . 0 9 1 1 0 N 4 4 6 6 A P U e T a S r s C o O 1 . 6 . 0 7 0 1 0 * * S ig . ( 2 . 0 0 0 . Su m o f 0 9 2 5 7 . 4 . 7 3 3 5 9 C r o s s C o v a r i 1 3 5 1 6 . 0 . 1 1 7 0 2 N 4 4 6 6 2 P P A C e f r S a ic C ra s O n o 1 . 0 . 5 0 7 0 8 * S i g . ( 2 . 0 1 9 . Su m o f 4 6 7 5 1 . 7 . 8 5 7 0 5 C r o s s C o v a r i 1 3 7 1 . 7 . 4 1 5 7 8 N 1 1 6 6 A P U e T aS r s C o O 1 . 5 . 0 7 0 8 0 * S ig . ( 2 . 0 1 9 . Su m o f 5 7 0 1 3 . 8 . 9 7 3 5 8 C r o s s C o v a r i 1 3 6 1 6 . 4 . 9 5 2 8 9 N 1 1 6 6 . P P C eS aC r s O o 1 . 0 . 6 0 1 0 3 S i g . ( 2 . 3 8 7 . Su m o f 8 3 4 6 . 0 . 0 0 0 0 0 C r o s s C o v a r i 2 1 8 2 . 0 . 0 0 0 0 0 N 4 4 A P U eT aS r s C o O 1 . 6 . 0 1 0 3 0 S ig . ( 2 . 3 8 7 . Su m o f 3 4 6 1 . 0 . 0 0 0 0 0 C r o s s C o v a r i 1 1 2 3 . 0 . 6 0 6 0 7 N 4 4 * * C * . C . o o r r r r 46 e e la la t io t io Summary We have investigated variables hypothesized to have an effect on whether or not a woman has inadequate finances. Using logistic regression, we found that employment history was not a good predictor of inadequate finances. The number of years a woman was employed was not a significant predictor, either alone or in a model with health status. Similarly, the interaction of whether or not a woman had ever been employed with health status was not significant in a loglinear analysis, and EVEREMPL was not a significant predictor in a logistic regression model with health status. Health status, however, was found to be a significant predictor of inadequate finances, and neither of the employment variables had much effect on the relationship when they were added separately to different logistic regressions. The effect of health status was also found to be significant in the loglinear analysis. Women on poor health were found to be about 1.4 times more likely than women in fair health and 3.6 times more likely than women in good health to have inadequate finances, regardless of whether or not either of the employment variables was included in the model. Multivariate analysis of variance was used to investigate the effect of inadequate finances on personal control and autonomy scores considered simultaneously. Financial status was significant predictor if = .06. Discriminant function analysis showed that personal control scores were more important than autonomy scores in discriminating between women with and without adequate finances. With the addition of ethnicity to the MANOVA model, financial situation could no longer be considered a significant predictor of the scores considered simultaneously. The interaction was not significant, although plots of marginal means suggested 47 that the autonomy scores of African American women were more strongly influenced by financial status than were those of Caucasian women. Of the variables and methods used with regard to inadequate finances, a simple bivariate logistic regression using health status with dummy variables seems the most desirable. 48