User manual SPSS MISSING VALUE ANALYSIS 7.5

[. . . ] SPSS Missing Value Analysis 7. 5 TM MaryAnn Hill / SPSS Inc. SPSS Inc. 233 South Wacker Drive, 11th Floor Chicago, IL 60606-6412 Tel: (312) 651-3000 Fax: (312) 651-3668 For more information about SPSS® software products, please visit our WWW site at http://www. spss. com or contact Marketing Department SPSS Inc. 233 South Wacker Drive, 11th Floor Chicago, IL 60606-6412 Tel: (312) 651-3000 Fax: (312) 651-3668 SPSS is a registered trademark and the other product names are the trademarks of SPSS Inc. No material describing such software may be produced or distributed without the written permission of the owners of the trademark and license rights in the software and the copyrights in the published materials. [. . . ] -. 3 . 7 . 9 1. 1 -. 9 -. 6 2. 3 1. 6 1. 8 . 4 1. 7 2. 0 -. 2 -1. 2 -. 8 1. 3 -3. 5 . 2 . 1 . 9 1. 4 -. 8 . 0 -16. 5 Missing Data 37 The most extreme t value is ­16. 5 for the grass item split into groups based on the pattern of sathobby (the respondents satisfaction with nonworking activities, or hobbies). In the full table with all layers displayed (not shown), we find that only eight cases have values of sathobby missing when grass is present; so we do not pursue this relation further. These variables belong to the same subset identified above and, if missing, tend to be missing together. The tests based on the pattern of rincom91 (respondent's income on the 1991­93 surveys) are more interesting. The t statistics are sizable for more than 20 items, including income91 (total family income from all sources). See Figure 2. 5 for boxplots of four of these distributions: age, education, income, and political views grouped by the pattern of rincom91. Figure 2. 5 90 Boxplots of age, education, family income, and political views grouped by the pattern of rincom91 25 25 8 7 20 20 6 15 15 70 50 POLVIEWS missing present INCOME91 5 4 3 2 EDUC 10 AGE 10 30 5 5 1 10 missing present 0 missing present 0 0 missing present RINCOM91 RINCOM91 RINCOM91 RINCOM91 The 34% of the respondents for whom rincom91 is missing tend to be older, have less education, have a lower total family income, and are more conservative (low values of polviews indicate liberal political views) than the 66% who do report rincom91. If rincom91 is crosstabulated against wrkstat (this variable's eight codes define job status), more than 80% of the respondents with missing rincom91 values report they are retired, in school, or a homemaker. 38 Chapter 2, Example 3 A Look at Multiple Missing Value Codes When the pattern variable has a few distinct categories, it is easy to use SPSS's boxplots to compare distributions of a measure within valid categories against those for one or more types of missing values. In the previous two-sample t tests, each pattern variable had to have at least 20% of its values missing. Lowering this cutoff to the default value of 5%, would add the big band and heavy metal music items as pattern variables. The two-sample t statistics for testing differences in average age for the groups formed by the two patterns are respectively, ­6. 95 and 6. 24. The average age for the 163 subjects with missing big band values is 37. 9 years; the average age for the 77 people with missing heavy metal values is 60. 4 years. The variables bigband and hvymetal have two missing value codes: 8 for don't know much and 9 for no answer. Figure 2. 6 and Figure 2. 7 examine the age distribution within each of these missing value categories. Figure 2. 6 100 582 591 709 530 651 894 704 1467 590 966 554 425 159 436 420 1349 People who "don't know much" about big band music are younger 80 1089 60 Age of Respondent 40 20 0 N= 252 544 268 216 52 151 12 Like very much Like it Mixed feelings Dislike very much Don't know much NA Dislike it Bigband Music In both Figure 2. 6 and Figure 2. 7, the age distribution of the few subjects with no answer (NA) differs from that for the don't know much group. Missing Data 39 Figure 2. 7 100 People who "don't know much" about heavy metal music are older 333 80 346 323 1349 1056 929 1096 1112 115 1464 41 60 1246 381 744 407 189 1139 1379 948 1478 956 1148 474 Age of Respondent 40 20 0 N= 45 116 175 365 717 67 10 Like very much Like it Mixed feelings Dislike very much Don't know much NA Dislike it Heavy Metal Music The Missing Value feature for crosstabulating pattern variables against categorical variables is useful for studying relationships among categories with different missing value codes. Here we define polviews (Where do you place yourself on a scale of political views ranging from Extremely liberal to Extremely conservative?) as a categorical variable and show how the categories relate to selected patterns. Indicator Variable Statistics ; Crosstabulation of categorical and indicator variables Omit variables missing less than 5% of cases The pattern variables include cappun (Do you favor or oppose the death penalty for persons convicted of murder?), letdie1 (If a person is in an advanced stage of a terminal illness, should doctors be allowed to end the patient's life if he and his family request it?), and the respondent's preference for big band and heavy metal music. 40 Chapter 2, Example 3 Overall, 7. 1% of the respondents answered that they don't know about capital punishment, but this response is spread unevenly across the seven categories. Almost 10% of the moderates say they don't know, while the percentages for those in the three liberal and three conservative categories are collectively closer to 5%. Among those who don't know (DK) where they belong on the political scale, 18. 8% also don't know where they stand regarding capital punishment (this is considerably higher than the overall 7. 1% rate). POLVIEWS Extremely conservative Slightly conservative Extremely liberal Slightly liberal Conservative Moderate Liberal Total Missing DK 39 81. 3 18. 8 . 0 25 52. 1 41. 7 6. 3 . 0 40 83. 3 12. 5 4. 2 35 72. 9 22. 9 4. 2 NA 6 66. 7 22. 2 11. 1 2 22. 2 44. 4 11. 1 22. 2 5 55. 6 11. 1 33. 3 5 55. 6 11. 1 33. 3 CAPPUN Present Missing LETDIE1 Present Missing BIGBAND Present Missing HVYMETAL Present Missing Count Percent % DK % NA Count Percent % NAP % DK % NA Count Percent % Don't know % NA Count Percent % Don't know % NA 1388 92. 5 7. 1 . 4 956 63. 7 32. 3 3. 7 . 3 1337 89. 1 10. 1 . 8 1423 94. 9 4. 5 . 7 29 96. 7 3. 3 . 0 22 73. 3 26. 7 . 0 . 0 25 83. 3 13. 3 3. 3 28 93. 3 3. 3 3. 3 152 93. 3 5. 5 1. 2 103 63. 2 35. 6 1. 2 . 0 145 89. 0 10. 4 . 6 160 98. 2 1. 2 . 6 184 95. 3 4. 7 . 0 125 64. 8 32. 1 3. 1 . 0 167 86. 5 13. 0 . 5 182 94. 3 5. 2 . 5 473 89. 8 9. 9 . 4 344 65. 3 29. 4 4. 9 . 4 465 88. 2 11. 6 . 2 508 96. 4 3. 6 . 0 238 96. 0 3. 6 . 4 161 64. 9 33. 1 2. 0 . 0 227 91. 5 8. 1 . 4 238 96. 0 3. 6 . 4 228 94. 6 5. 4 . 0 148 61. 4 34. 4 4. 1 . 0 224 92. 9 6. 2 . 8 229 95. 0 4. 6 . 4 39 95. 1 4. 9 . 0 26 63. 4 31. 7 4. 9 . 0 39 95. 1 4. 9 . 0 38 92. 7 7. 3 . 0 In the Extremely liberal and Liberal categories, there are fewer folks (0. 0% and 1. 2%) who are uncertain (DK) about euthanasia than there are in the conservative groups (4. 1% and 4. 9%). With respect to big band music, there is a linear decrease in the percentage answering don't know across the seven political categories (13. 3%, 10. 4%, 13%, 11. 6%, 8. 1%, 6. 2%, and 4. 9%)--conservatives are less apt to report that they don't know about big band music. The opposite is true for heavy metal music. Missing Data 41 Example 4: Estimating Means, Standard Deviations, Covariances, and Correlations In the Missing Value procedure, the user can choose to estimate means, standard deviations, covariances, and correlations using listwise (complete cases only), pairwise, EM, and/or regression methods. The Missing Value procedure also provides all values estimates of means and standard deviations plus several options for the EM and regression methods. When more than one method is requested, the estimates of the means are displayed in one summary panel, and the estimates of the standard deviations are displayed in another (see p. Over the years, many software users approached the missing data problem by using a pairwise complete method to compute a covariance or correlation matrix and then using this matrix as input for, say, a factor analysis. [. . . ] This specification limits the total number of predictors in the analysis. The analysis uses the stepwise selected n best predictors, entered in accordance with the tolerance. If n = 0 , it is equivalent to replacing each variable with its mean. FLIMIT=n NPREDICTORS=n 76 MVA ADDTYPE Type of distribution from which the error term is randomly drawn. Random errors can be added to the regression estimates before the means, correlations, and covariances are calculated. [. . . ]