HOW TO USE DAG_Stat 


Input
Other Controls
Results
Statistics for Diagnostic Tests
Statistics for Interrater Agreement
Other Indices
Other Output
Notes
INPUT
 The input to DAG_Stat are the four cell counts. In most cases, you will have produced these counts using one of the general-purpose statistical packages such as SPSS, SAS, Stata etc. Excel itself has limited statistical capablity usually accessed via the "Add-Ins..." command on the "Tools" menu. There are other packages available that work with Excel that crosstabulate data. Of course, the input data might come from published research or any other source.

It is possible to enter your own name for the disease or condition being diagnosed by replacing "Criterion" and, similarly the name of the Test. This may be useful to help you identify output, but it has no effect on any statistical calculations.

The probability coverage of the confidence intervals may also be changed as below

Normally 95% confidence intervals are reported, but any value greater than 0 and less than 100% may be entered.

Other controls

Four buttons activate macros that perform a number of functions.

Clear removes all four cell entries.

Calculate forces recalculation of the results. This button is useful with older, slower computers. Calculation may be set to manual (on the Options command in Tools menu). This prevents Excel from recalculating statistics as each cell is entered. After all four values are input, the Calculate button is then pressed. This option is unnecessary on modern, fast machines.

% displays appropriate Diagnosic Test indices in percent form.

0.0 displays appropriate Diagnosic Test indices in percent form.

RESULTS

Statistics for Diagnostic Tests

DAG_Stat produces a comprehensive range of indices that are widely used in the assessment of the performance of diagnostic tests. In addition, a number of less well known indices such as the sensitivity of a test working at a random level of performance and an index of senstivity that adjusts for this (the quality index) are produced. Some indices, such as Youden's index have been included principally to allow comparison with published results rather than because they are recommended statistics.

Each index is briefly explained in a pop-up comment that will appear when you place the mouse over the at the top righthand corner of the index name. (N.B. Depending on how your copy of Excel is configured, this feature may need to be activated.)

Interrater Agreement Statistics
DAG_Stat calculates Cohen's Kappa, associated standard error and test of the hypothesis that kappa differs from zero. It also shows the observed agreement and that expected by chance alone. A standardized description of the magnitute of agreement is provided. More information about the nature of agreement (or disagreement) can be gained by considering agreement for positive and negative judgements separately. This is acheived through indices of positive and negative agreement. Difference in the judged prevalence of an outcome and the relative frequency of positive versus negative outcomes can affect Kappa. Byrt's indices and adjustments can assist in identifying these effects. McNemar's test may also be useful in this regard.

Other Indices and Tests

A large number of other indices have been proposed as measures of agreement or have theoretical models that are relevant to the performance of tests or judges. DAG_Stat calculates a wide range of these measures. It should be noted that many of these indices have substantial limitations or are based on models that may not be relevant to particular situations. These statistics are provided principally to allow comparison with published results.

The chi-square test (both Pearson and likelihood ratio) is calculated as a general test of independence of the test and criterion. McNemar's test in this application tests for differences in judged prevalence of the disorder being studied.

Other Output

A table of percentages in each cell is provided beside the input cells.

In addition to output in tabular form, test statistics are also produced in an abbreviated form for copying and pasting into reports. The number of decimal places used in these summaries is specified at the top of the sheet.

"Sensitivity = 0.85 (95% CI: 0.82 - 0.88)"

"Sensitivity = 0.8538 (95% CI: 0.8216 - 0.8822)"

Notes
Some indices cannot be calculated when one or more cell entries are zero. A warning is displayed under these conditions. Low expected frequencies may invalidate the chi-square test. Accordingly, the number of cells with expected frequencies below 5 and below 1 are reported.

In the original version of DAG_Stat the likelihood ratio for a negative test was report as the inverse of the conventional value so that it had the same interpretation as the ratio for a positive test (it's what you'd get if you swapped positive and negative around).  This idea did not take on and created confusion.  So the current version now reports the conventional LR- index, as well as its inverse, both clearly marked.

As downloaded, each sheet is protected from modification. Only the shaded cells on the Statistics worksheet may be changed. If you wish to modify the sheet, choose the "Unprotect Sheet..." option from the "Protection" command on the "Tools" menu.

Go to Top

Back to Main DAG_Stat Page

Home

Last updated January 2011.