Journal of Aflectiue Disorders, 20 (1990) 249-250 Elsevier
249
JAD 00763
Climatic variables and admissions for mania: a reanalysis David F. Peck Area Clinical Psychology Service (Highland Health Board), Craig Phadrig Hospital, Inverness IV3 6PJ. U.K (Received 6 April 1990) (Revision received 20 August 1990) (Accepted 21 August 1990)
Summary Differencing and cross-correlations are useful and straightforward methods for dealing with relationships between variables measured over short periods of time. Without such transformations of time-series data, spurious relationships may emerge and the true nature of underlying relationships may be obscured. This reanalysis of data from Camey et al. (1989) suggests that there is a strong relationship between amount of sunshine in a month and admissions for mania one month later.
Key words: Climate;
Mania;
Differencing;
Cross-correlations
In a recent paper Carney et al. (1989) reported high correlations between admissions for mania and two climatic variables, namely hours of sunshine per month and day length. Data for these three variables were collated from January to December for a 5 year period from 1980 to 1984, and mean hours of sunshine, average day length and total admissions were calculated. Several standard product moment correlations between admissions and climatic variables were then conducted; for example, the correlation between admissions and sunshine over the 5 years was +0.70. Several commentators (e.g., Gibbons and
Address for correspondence: Mr. D.F. Peck, Area Clinical Psychology Service (Highland Health Board), Craig Phadrig Hospital, Inverness IV3 6PJ, U.K. 0165-0327/90/$03.50
0 1990 Elsevier Science Publishers
Davis, 1984) have pointed out that such time-series data are notoriously difficult to analyse and to interpret, and that the use of standard correlational methods (e.g., product moment correlations) may give rise to misleading conclusions and may obscure underlying relationships. Furthermore one might expect that if there was a relationship between day length and admissions (for example), it would be unlikely that the effect would occur immediately; it may take time for any changes in day length to produce mood change, and for mood changes to become apparent in admission statistics. Thus correlating admissions and climate during the same month may not be the most appropriate way to investigate the relationship between them. Plant et al. (1988) have outlined more statistically valid methods to deal with the analysis of
B.V. (Biomedical
Division)
250
such time-series data when the data sets are short and when a delayed effect might be anticipated, as in the study by Carney and colleagues. Full details are available in that paper. The methods involve the use of differencing and cross-correlations. In differencing, the value of each observation in each variable is subtracted from the value preceding it; the two sets of differenced data can then be correlated. One is no longer examining the relationship between the variables as they change over time. but rather whether the variables change in the same direction and to a similar degree; in the present context, is an increase (or decrease) in a climatic variable accompanied by a similar increase (or decrease) in admissions for mania? In other words, if there is a large increase in sunshine from one month to the next, this should be accompanied by a proportional increase in admissions. Cross-correlations enable one to analyse the relationship between two variables, with one variable delayed in time. A delay of one time unit is referred to as lag(l). In the present context, climatic variables in January would be correlated with admissions in February; climatic variables in February with admissions in March, and so on. These methods were used to reanalyse the data of Carney and colleagues. The data were taken from their Fig. 2. which is reproduced here. The data were reanalysed using the Minitab statistical package and the results can be seen in Table 1, along with the original correlations.
MONTHS
1. Reproduced from Carney et al. (1989), with permission of the authors, editors and publishers.
1
CORRELATIONS OF WITH ADMISSIONS
SUNSHINE
Sunshme Differenced Admissions
AND
DAY
LENGTH
Day length
data +0.146
(NS) DrJjerenced and lug(I) dam Admissions + 0.77
(P < 0.01) Original data from Gurney et ul. (I 989) Admissions +0.70 (P i 0.05)
+ 0.327 (NS) f 0.445
(NS) + 0.70
( P i 0.02)
Differenced data and both differenced and lagged data compared with original data from Carney et al. (1989).
were
It can be seen that differencing markedly reduces the size of the correlation coefficients, and that the relationship between climatic variables and admissions is no longer significant. However, when a lag of one month is also introduced, a strong relationship emerges between mean sunshine hours and admissions 1 month later (1. = +0.77, P < 0.01). Such a relationship makes a great deal of intuitive sense. This correlation is not only statistically valid, but is slightly greater than that reported by Carney and colleagues. Causality cannot be assumed, but longitudinal studies of individual patients may further elucidate the nature of this relationship.
References
1'3
Fig.
TABLE
kind
Carney, P.A., Fitzgerald, C.T. and Monaghan, C. (1989) Seasonal variations in mania. In: C. Thompson and T. Silverstone (Eds.), Seasonal Affective Disorder. CNS (Clinical Neuroscience) Publishers. London, pp. 19-27. Gibbons, R.D. and Davis, J.M. (1984) The price of beer and the salaries of priests: analysis and display of longitudinal psychiatric data. Arch. Gen. Psychiatry 41. 1183-1184. Plant, M.A., Peck, D.F. and Duffy, J.C. (1988) Trends in the use and misuse of alcohol and other psychoactive drugs in the United Kingdom: some perplexing connections. Br. J. Addict. 83, 943-947.