I agree with previous responses to this query but should add that if you are in SAS already you will find that the new MI procedure does all that NORM does, and more. I've just started to use it and find it really easy. I've detected a couple of bugs that I have jsut reported to SAS and I am sure will be fixed soon (see copy below). I'd be interested to know if any other list members have been trying out MI.
On a more general matter I don't think there is any need to worry about distributional assumptions when one only has spotty missing data, as here. A practical problem arises, though when one has a classification (eg red, green, blue) and imputes from two dummies. Obviously we needto something when an imputation gives red and green. I have fudging code for this. Does anyone have an elegant solution? ---------------------------------------------------------------------------- -------------------- To SAS technical support Thanks for your help earlier this week about the new MI procedure. I managed to solve the problem I mentioned to you and I append a note about it that I would be grateful if your could pass on to the developper. Meanwhile I have come accross another problem which I document below that I'd also like you to pass on. Problem 1. If MI with normal imputation and MCMC options is aksed to impute a variable with all values the same then it crashes with a zero divide error and we get no output. A patch to get over this should be easy as it would be obvious what any missing values for this should be (unless we want to be paranoid and worry about our data set not having seen other possible values that might have happened). Problem 2 This is more complicated. It concerns the maximum and minimum and rounding options. I have got some funny results from a data set where after MI some of my variables have had perfectly acceptable non-missing values replaced with values outside the allowed range. It happened to the last two variables in my list, and when I moved the variables to the ehad of the list they were OK. I attach the program and the two sets of output tables of results. I could also make the data available to the developper if this would help. Another problem is that my documentation disc has some help on MI, but it doesn't give much detail of how rounding and limits are handled. Am I right in thinking that what happens is that the data are imputed, rounded, and then checked to see if outside limits, and then a further sample taken if we are outside range? When I have programmed this myself I have always just put the extreme ones into the topmost category. Is what MI is doing better? I would expect these two methods to have opposite biases, so maybe some combination would be good, bu tI've not really thought it through fully. To whoever is writing MI - well done, it is really easy to use and will make my life much easier. Thank you. Prof Gillian Raab Applied Statistics Group School of Mathematics and Statistics Napier University, Sighthill (room 419a), Edinburgh EH11 4BN tel 0131 455 3532 fax 0131 455 3485 -------------- next part -------------- References Bagust A, Roberts B, Haycox A, Barrow S (1999), Cost of health care. The additional cost of obesity to the health service and the potential for resource savings from effective interventions, The European Journal of Public Health 9: 258-264 Ref ID: 121 Reprint: Not in File Blakely TA, Woodward AJ (2000), Ecological effects in multi-level studies, Journal of Epidemiology and Community Health 54: 367-374 Ref ID: 5 Reprint: Not in File Blane D, Power C, Bartley M (1996), Illness behaviour and the measurement of class differentials in morbidity, Journal of the Royal Statistical Society.Series A.Statistics in society 159: 77-92 Ref ID: 102 Reprint: In File Chen Y, Rennie DC, Reeder BA (1995), Age-related association between body mass index and blood pressure: the Humboldt Study., Int J Obes Relat Metab Disord 19: 825-831 Ref ID: 106 Reprint: Not in File Curtis SE (1990), Use of survey data and small area statistics to assess the link between individual morbidity and neighbourhood deprivation., J Epidemiol Community Health 44: 62-68 Ref ID: 108 Reprint: Not in File Davey Smith G, Brunner E (1997), Socio-economic differentials in health: the role of nutrition, Proceedings of the Nutrition Society 56: 75-90 Ref ID: 57 Reprint: In File Davies H, Joshi H, Clarke L (1997), Is it cash that the deprived are short of ?, Journal of the Royal Statistical Society.Series A.Statistics in society 160: 107-126 Ref ID: 71 Reprint: In File Diez-Roux AV (1998), Bringing context back into epidemiology: Variables and fallacies in multilevel analysis, American Journal of Public Health 88: 216-222 Ref ID: 20 Reprint: In File Diez-Roux AV, Nieto FJ, Caulfield L, Tyroler HA, Watson RL, Szklo M (1999), Neighbourhood differences in diet: the Atherosclerosis Risk in Communities (ARIC) Study, Journal of Epidemiology and Community Health 53: 55-63 Ref ID: 15 Reprint: Not in File Dubios L, Girard M, Bergeron N (2000), The choice of a diet quality indicator to evaluate the nutritional health of populations, Public Health Nutrition 3: 357-365 Ref ID: 123 Reprint: Not in File Duncan C, Jones K, Moon G (1993), DO PLACES MATTER - A MULTILEVEL ANALYSIS OF REGIONAL VARIATIONS IN HEALTH-RELATED BEHAVIOR IN BRITAIN, Social Science & Medicine 37: 725-733 Ref ID: 44 Reprint: In File Duncan C, Jones K, Moon G (1996), Health-related behaviour in context: A multilevel modelling approach, Social Science & Medicine 42: 817-830 Ref ID: 33 Reprint: In File Duncan C, Jones K, Moon G (1998), Context, composition and heterogeneity: Using multilevel models in health research, Social Science & Medicine 46: 97-117 Ref ID: 23 Reprint: In File Duncan C, Jones K, Moon G (1999), Smoking and deprivation: are there neighbourhood effects?, Social Science & Medicine 48: 497-505 Ref ID: 12 Reprint: In File Ecob R (1996), A multilevel modelling approach to examining the effects of area of residence on health and functioning, Journal of the Royal Statistical Society Series A-Statistics in Society 159: 61-75 Ref ID: 35 Reprint: In File Ecob R, Macintyre S (2000), Small area variations in health related behaviours; do these depend on the behaviour itself, its measurement, or on personal characteristics?, Health & Place 6: 261-274 Ref ID: 1 Reprint: In File Gould M, Jones K, Moon G (1997), The scope of multilevel models, Environment and Planning A 29: 581-584 Ref ID: 29 Reprint: Not in File Haynes R, Gale S, Lovett A, Bentham G (1996), Unemployment rate as an updatable health needs indicator for small areas, Journal of Public Health Medicine 18: 27-32 Ref ID: 126 Reprint: Not in File Haynes R, Lovett A, Reading R, Langford I, Gault S (1999), Use of homogeneous social areas for ecological analyses. A study of accident rates in pre-school children, The European Journal of Public Health 9: 218-222 Ref ID: 128 Reprint: Not in File Hupkens C, Knibble R, Drop M (2000), Social class differences in food consumption. The explanatory value of permissiveness and health and cost considerations, The European Journal of Public Health 10: 108-113 Ref ID: 127 Reprint: Not in File Macintyre S, Maciver S, Sooman A (1993), Area, class and health - should we be focusing on places of people? Journal of social policy 22: 213-234 Ref ID: 124 Reprint: Not in File Mackenbach JP (1992), Socio-economic health differences in the Netherlands : a review of recent empirical findings, Social Science & Medicine 34: 213-226 Ref ID: 89 Reprint: Not in File Manor O, Matthews S, Power C (1997), Comparing measures of health inequality, Social Science & Medicine 45: 761-771 Ref ID: 28 Reprint: Not in File Manor O, Matthews S, Power C (2000), Dichotomous or categorical response? Analysing self-rated health and lifetime social class, International Journal of Epidemiology 29: 149-157 Ref ID: 6 Reprint: Not in File OReilly D, Stevenson M (1998), The two communities in Northern Ireland: deprivation and ill health, J Public Health Med 20: 161-168 Ref ID: 114 Reprint: Not in File Purdon S, Korovessis C, Thomas R, Hedges B (2001), The effect of area of residence on health, preprint Ref ID: 125 Reprint: In File Saul C, Payne N (1999), How does the prevalence of specific morbidities compare with measures of socio-economic status at small area level?, J Public Health Med 21: 340-347 Ref ID: 115 Reprint: Not in File Smith GD, Hart C, Watt G, Hole D, Hawthorne V (1998), Individual social class, area-based deprivation, cardiovascular disease risk factors, and mortality : the Renfrew and Paisley study [Full text available, price (Pounds)9.40], Journal of Epidemiology and Community Health 52: 399-405 Ref ID: 93 Reprint: Not in File Twigg L, Moon G, Jones K (2000), Predicting small-area health-related behaviour: a comparison of smoking and drinking indicators, Social Science & Medicine 50: 1109-1120 Ref ID: 7 Reprint: Not in File Wagstaff A, Paci P, VanDoorslaer E (1991), On the measurement of inequalities in health, Social Science & Medicine 33: 545-557 Ref ID: 97 Reprint: Not in File Woodward M (1996), Small area statistics as markers for personal social status in the Scottish heart health study, Journal of Epidemiology and Community Health 50: 570-576 Ref ID: 101 Reprint: Not in File 3 -------------- next part -------------- A non-text attachment was scrubbed... Name: forsas2.htm Type: application/octet-stream Size: 13621 bytes Desc: not available Url : http://lists.utsouthwestern.edu/pipermail/impute/attachments/20010426/d702aaa5/forsas2.obj -------------- next part -------------- A non-text attachment was scrubbed... Name: forsas1.htm Type: application/octet-stream Size: 18411 bytes Desc: not available Url : http://lists.utsouthwestern.edu/pipermail/impute/attachments/20010426/d702aaa5/forsas1.obj -------------- next part -------------- A non-text attachment was scrubbed... Name: imptest_for_sasinst.sas Type: application/octet-stream Size: 3621 bytes Desc: not available Url : http://lists.utsouthwestern.edu/pipermail/impute/attachments/20010426/d702aaa5/imptest_for_sasinst.obj
