[EMAIL PROTECTED] (Pingu) wrote in news:[EMAIL PROTECTED]:
> 12 subjects took part. Each subject had their own 10 name phonebook. > Each subject had to dial every name in their phonebook (randomly > presented) under four different conditions (also randomised): > > 1) Voice dialling (VD) + quiet > 2) VD + noise > 3) TTS dialling + quiet > 4) TTS dialling + noise [snip] > My tutors say that i cant produce these t tests without first running > a 2 way repeated measures ANOVA on all the data together to see if > there is a general effect first, only then can i parcel the analyses > into those for quiet and those for noisy. To quote them: "Start the > analyses with 2-way ANOVAs of the overall data from each > quiet/noisy/VAD/TST set and then go on to do the separate t-tests" > > I just cant get my head round this. I guess my main problem is that i > dont know what an ANOVA does, and consequently when i do them i really > have no idea what the output means. To me it just makes no sense to > analyse the data with the noisy and quiet conditions taken together - > it is not interesting to know if there is an effect between VD Quiet > and TST noisy for example, because it is not a fair comparison. So > what do these tests actually tell me? Are my tutors misunderstanding > the point of my experiements? Your tutors are correct. What an ANOVA does in your case is estimate the parameters of a model: S = a1*M + a2*N + a3*M*N + E Where S is the dialing speed, M is the type of dialing method, N is the noise level, M*N is an "interaction term" and E is the "error term." Essentially this model says that there are four influences on dialing speed: 1) The effect of the dialing method chosen 2) The effect of the noise level 3) The effect of interaction between the dialing method and the noise level; if there is any interaction, this means that the effect of dialing method choice under quiet conditions is different from its effect under noisy conditions. 4) All the other possible influences that you haven't included in the model, such as the user's voice characteristics, hearing ability, reaction time, and other sources of individual variation. This is technically called "error" which is somewhat misleading. An ANOVA will give you an overall test of significance which tests the null hypothesis that a model containing only E fits the data just as well as one including M and N. It will also give you estimates for a1, a2, and a3, along with tests of the null hypotheses that each of them are zero. If, for example, a1 and a3 don't significantly differ from zero, then you can conclude that the dialing method doesn't make a difference. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
