Non-Response Bias

 < Day Day Up > 



As in the case of any survey research, there is the potential for those individuals returning questionnaires not to accurately represent the population studied (i.e., non-response bias). In order to examine the possibility of non-response bias, the extreme quartiles of the sample, when ordered by questionnaire return date, were studied (Armstrong & Overton, 1977). Specifically, the upper quartile was used to represent or simulate non-respondents and the lower quartile the respondents. The simulated non-respondents and respondents were then compared for differences in their demographics and summated measures. No meaningful differences between the simulated non-respondents and respondents imply an absence of non-response bias in the sample (Armstrong & Overton, 1977). The demographics measured as continuous variables (i.e., age of the respondent and years of employment with the organization) were compared using t-tests. The discrete demographic variables (i.e., gender, educational level, and position in the hospital) were compared between these groups using a chi-square goodness of fit test. Meaningful differences between the simulated non-respondents and respondents, using a 5% significance level, were found for educational level and position in the hospital. The specific values of the test statistics were: age t=-0.31, length of employment at the hospital t=0.26, gender chi-square(6)=6.87, educational level chi-square(4)=24.56, and position in the hospital chi-square(4)=9.65. Inspection of the educational level of the simulated non-respondents and respondents indicated that the primary difference between these groups was found in the respondents with graduate degrees. A full 10% of the simulated non-respondents possessed a graduate degree, while 0% of the respondent group held a graduate degree. Given the link between educational level in the hospital, particularly holding a graduate degree, and position, this likely influenced the meaningful difference between these groups for the position variable as well.

Because of the significant differences for these two demographic variables, additional tests were performed. The constructs measured using multiple questionnaire items were captured by summing the values of the individual items. These summated measures were then analyzed using multiple analysis of variance along with the continuous, single item measure of the degree of system use. The results of this analysis showed no significant differences either individually or as a group. The F-statistic for each measure were: computer self-efficacy F(1,162)=0.18, outcome expectancy F(1,162)=0.30, past computer experience F(1,162)=0.38, computer staff support F(1,162)=0.03, ease of system use F(1,162)=0.29, degree of system use F(1,162)=0.39, and organizational commitment F(1,162)=1.06. Wilks' Lambda was used to measure the group significance of these measures. Its value was 0.98 producing an F-statistic with seven and 156 degrees of freedom equal to 0.38. Thus, the measures used in the analysis did not differ in a significant way between the simulated non-respondents and respondents. It is these measures and their interrelationships that are at the heart of the analysis. Since these measures did not demonstrate significant differences between the simulated non-respondents and respondents, non-response bias should not present a serious problem in the sample or the study.



 < Day Day Up > 



Advanced Topics in End User Computing (Vol. 3)
Advanced Topics in End User Computing, Vol. 3
ISBN: 1591402573
EAN: 2147483647
Year: 2003
Pages: 191

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net