Odtnhj

I want to ask about linear mixed models.

The significance of the fixed variable is changed with a random structure.
For example, suppose there are 5 variables:
RT(response variable), covariate variable1(C.V.1), C.V.2, I.V.1, I.V.2.
all variables are within-subject variables excepting RT.

What I want to know is the interaction of I.V.2 and I.V.2.

In this situation, I set the two models using lmer().
First is that:

m1 <- lmer(rt ~ C.V.1 + C.V.2 + I.V.1*I.V.2 + (1+C.V.1 + C.V.2 + I.V.1*I.V.2|subject) + (1|word))

and second is that:

m2 <- lmer(rt ~ C.V.1 + C.V.2 + I.V.1*I.V.2 + (1+ I.V.1*I.V.2|subject) + (1|word))

when I analyzed this two models, the significance of fixed variable is different between the two models.
For example, the interaction of I.V.1 and I.V.2 is significant in m1, but not in m2.

I know setting the subject intercept means the responses would be different from each subject and setting the subject slope for I.V.1 means the effect of I.V.1 would be different from each subject.

But I don't know the relationship between fixed effects and random effects.
What is the meaning that considering random effects?
Can I interpret the result of the estimate of fixed variable as a coefficient when controlling the effect of other random effects like covariate variable?

And why the significance of fixed effect is changed with a random structure like the above two models?

Thank you for reading and I hope anyone would explain me these.