A review of commonly applied statistics in JAN

    loading  Checking for direct PDF access through Ovid

Excerpt

It is well established that nurses and midwives champion evidence‐based practice as demonstrated by both the attention towards reducing the research‐to‐practice gap (Brant 2015) and proliferation of published nurse‐led and midwife‐led research studies. In keeping with this culture of enquiry, we carried out an examination of the type and frequency of commonly applied inferential statistical techniques applied to research studies published in JAN. The findings should provide researchers, educators and authors of healthcare statistic books with an overview of the types of statistical tests used in research studies.
Two decades ago, Anthony (1996) reported findings for a somewhat similar descriptive review of statistics published in JAN, which served as a suitable baseline from which we could compare and contrast our results. To facilitate a meaningful comparison, we planned our review to be as analogous as possible with Anthony's study. First, we conducted our review over the same time span, that is, six months of journal issues. To do this, we examined JAN issues June–November 2015 inclusive. Second, we focused on the identification of inferential (not descriptive) statistics. Both reviews yielded approximately the same number of included articles, 45 (Anthony's review) and 46 (out of 108 research articles) for this review.
Table 1 shows that 16 different inferential statistical techniques were applied 111 times (66 parametric) and (45 non‐parametric) in 46 included research studies. The Chi‐square statistic ranked first as the most frequently applied inferential statistic. It was applied in 25 (54%) articles. Table 2 shows that these 25 studies used Chi‐square 28 times in three different ways: the comparison of proportions (21 studies); goodness of fit testing (6 studies); and testing for trends in proportions (1 study).
Table 1 shows that the Student's independent t‐test ranked as the second most frequently applied technique, used in 18 (39%) studies and is the most frequently applied parametric technique. More complex statistical techniques were also evident in the studies examined shown in Table 1. For example, regression analysis (ranked as third most frequency used statistic) featured in 17 (37%) studies. Analysis of variance techniques such as ANOVA (independent comparison of several means) was applied in seven (15%) studies, ANOVA (with repeated measure) five (11%) studies, ANCOVA, analysis of co‐variance three (7%) and MANOVA (Multivariate analysis of variance with several dependent variables) in two (4%) studies. When combined, Multiple Variable Analyses (MVA) techniques such as ANOVA (independent and repeated measure), ANCOVA and MANOVA amounted to 17 (37%) making it jointly ranked third with regression techniques, which was also found in 17 (37%) articles. Pearson correlation occurred in nine (20%) studies and ranked fourth in popularity.
Despite the popularity of the Chi‐square technique, a greater number of parametric techniques vs. non‐parametric methods were applied. For example, Table 1 (column 2) shows that eight parametric tests were used more frequently (66 times) compared with the eight non‐parametric tests (45 times) (Column 5). With reference to Anthony's (1996) descriptive review of 45 articles published in JAN, he identified 21 different statistics: Chi‐square (14; 31·1%), Mann–Whitney (9; 20%), ANOVA (7; 15·6%) Cronbach's alpha (7; 15·6%), Student's t‐test (6; 13·3%), factor analysis (4; 8·9%), Kruskall–Wallis (3; 6·7%), Spearman's rank (3; 6·7%), Pearson's (3; 6·7%), Student's t‐test (paired) (2; 4·4%) and Kappa (k) (2; 4·4%). In addition, the following list of tests were identified by Anthony in one (2%) of 45 articles: Fisher's exact test, test–retest, sign test, page rank test, Spearman–Brown, discriminant analysis, Jaccard index, multiple regression. Finally, Chi‐square for trend was not used.
    loading  Loading Related Articles