Ritual versus logic in significance testing in communication research

Thomas Steinfatt

Research output: Contribution to journalArticle

3 Citations (Scopus)

Abstract

Statistical tests of significance tend to produce Type I error probabilities far in excess of the reported “alphas” when more than one significance test is conducted on a data set. Since this condition is almost always the case in communication research, most reported alpha levels are little more than fiction as they relate to the occurrence of the Type I errors they are supposed to index and monitor. The calculation of an alpha percentage (α%) allows an estimation of the number of Type I errors being reported as “significant,” suggests a rational basis for where to set the nominal alpha level for a given data set in order to obtain any specified ratio of fictional to “real” results, and provides a general method of controlling Type I error rates since it is applicable across all tests of significance.

Original languageEnglish (US)
Pages (from-to)90-93
Number of pages4
JournalCommunication Research Reports
Volume7
Issue number2
DOIs
StatePublished - Dec 1 1990

Fingerprint

communication research
religious behavior
Communication
Testing
Statistical tests
significance test
statistical test

ASJC Scopus subject areas

  • Communication

Cite this

Ritual versus logic in significance testing in communication research. / Steinfatt, Thomas.

In: Communication Research Reports, Vol. 7, No. 2, 01.12.1990, p. 90-93.

Research output: Contribution to journalArticle

@article{98ce944b28164cbca0328d825daea836,
title = "Ritual versus logic in significance testing in communication research",
abstract = "Statistical tests of significance tend to produce Type I error probabilities far in excess of the reported “alphas” when more than one significance test is conducted on a data set. Since this condition is almost always the case in communication research, most reported alpha levels are little more than fiction as they relate to the occurrence of the Type I errors they are supposed to index and monitor. The calculation of an alpha percentage (α{\%}) allows an estimation of the number of Type I errors being reported as “significant,” suggests a rational basis for where to set the nominal alpha level for a given data set in order to obtain any specified ratio of fictional to “real” results, and provides a general method of controlling Type I error rates since it is applicable across all tests of significance.",
author = "Thomas Steinfatt",
year = "1990",
month = "12",
day = "1",
doi = "10.1080/08824099009359860",
language = "English (US)",
volume = "7",
pages = "90--93",
journal = "Communication Research Reports",
issn = "0882-4096",
publisher = "Routledge",
number = "2",

}

TY - JOUR

T1 - Ritual versus logic in significance testing in communication research

AU - Steinfatt, Thomas

PY - 1990/12/1

Y1 - 1990/12/1

N2 - Statistical tests of significance tend to produce Type I error probabilities far in excess of the reported “alphas” when more than one significance test is conducted on a data set. Since this condition is almost always the case in communication research, most reported alpha levels are little more than fiction as they relate to the occurrence of the Type I errors they are supposed to index and monitor. The calculation of an alpha percentage (α%) allows an estimation of the number of Type I errors being reported as “significant,” suggests a rational basis for where to set the nominal alpha level for a given data set in order to obtain any specified ratio of fictional to “real” results, and provides a general method of controlling Type I error rates since it is applicable across all tests of significance.

AB - Statistical tests of significance tend to produce Type I error probabilities far in excess of the reported “alphas” when more than one significance test is conducted on a data set. Since this condition is almost always the case in communication research, most reported alpha levels are little more than fiction as they relate to the occurrence of the Type I errors they are supposed to index and monitor. The calculation of an alpha percentage (α%) allows an estimation of the number of Type I errors being reported as “significant,” suggests a rational basis for where to set the nominal alpha level for a given data set in order to obtain any specified ratio of fictional to “real” results, and provides a general method of controlling Type I error rates since it is applicable across all tests of significance.

UR - http://www.scopus.com/inward/record.url?scp=0009890863&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0009890863&partnerID=8YFLogxK

U2 - 10.1080/08824099009359860

DO - 10.1080/08824099009359860

M3 - Article

AN - SCOPUS:0009890863

VL - 7

SP - 90

EP - 93

JO - Communication Research Reports

JF - Communication Research Reports

SN - 0882-4096

IS - 2

ER -